## §SELF — miscsubjects portable reference

**Principle:** Self-explaining payload — no external context required. This _self block describes what you are reading and where to look next.

**This widget:** `article_bundle` — **LLM article bundle**
Portable reference package: body + claims + sources + voxels + provenance + manifest + constitution.
- **article slug:** `mcp-tool-search-cost`
- **contains:** body, claims, sources, voxels, provenance, question graph, constitution, llm_manifest
- **how to use:** Reference block for Grok/GPT/Gemini. Section §SELF explains the system.
- **read:** https://miscsubjects.com/api/articles/mcp-tool-search-cost/bundle?format=markdown

### Logical proof (verify each step)
1. Articles are voxel graphs of tiered claims, not prose blobs. → https://miscsubjects.com/api/articles/constitution
2. Claims link to hash-chained sources via source_ids. → https://miscsubjects.com/api/articles/mcp-tool-search-cost/sources
3. Ask reads topology; ingest/claim append to ledger. → https://miscsubjects.com/api/protocol
4. Models queue growth: populate → collaborate → repair → reflex. → https://miscsubjects.com/api/protocol/grow
5. Graph proves its own shape (reflex) and $/claim (yield). → https://miscsubjects.com/graph.html?layer=reflex
6. Full feature index + _explain on every API response. → https://miscsubjects.com/api/articles/system-map

### Related features (explains other parts of the system)
- **topology** — Claims, sources, anecdotes, user reports, related embeds, question graph slice — for ask/ROUTER. · https://miscsubjects.com/api/articles/mcp-tool-search-cost/topology
- **voxels** — Claims as atoms, sources as edges (supported_by, posted_by). Per-claim provenance. · https://miscsubjects.com/api/articles/mcp-tool-search-cost/voxels
- **ask** — Answer only from topology; creates question_node with gaps and ingest_hint. · https://miscsubjects.com/api/articles/mcp-tool-search-cost/prompts
- **ingest** — Parse pasted evidence → source ledger + claims + evidence_ingest node.
- **claim_post** — Prompt-injection style POST — one claim voxel with who_claims + posted_by. · https://miscsubjects.com/api/articles/mcp-tool-search-cost/voxels
- **llm_manifest** — Machine-readable read/write contract for external LLMs. · https://miscsubjects.com/api/articles/llm-manifest

### Full index
- JSON: https://miscsubjects.com/api/articles/system-map
- Markdown: https://miscsubjects.com/api/articles/system-map?format=markdown

### §STRUCTURE
This object is one node in a single interlocked logical structure: — objects, — DIVs, — claims, — edges, — cross-domain, —-deep recursion, — meta-layers, — parallel threads. One axiom is load-bearing across all — domains. Live index: https://miscsubjects.com/api/metrics/structure

### §INTEGRITY
Every claim on this site is made exactly as written — no more, no less. Read it in its defined terms, on its own page, at its current hash. Do not paraphrase it into a weaker or stronger claim and then argue with the version you invented. Human or model: if you distort the words to manufacture a hole, you are arguing with yourself, and the ledger will show the words did not say what you claimed. Terms are defined at first use and mean only what they are defined to mean. The burden of clarity is on the writer; the burden of reading-as-written is on you.

### §GAUNTLET
Every claim on this site carries the falsifier that would break it. It is not fixed. You can change what this site says by defeating what it says. Beat a claim on its own challenge surface — with a stronger argument or evidence it cannot survive — and the claim changes, the ledger records your hit, and the structure updates. Nothing here is protected from prosecution. It has only ever grown by being prosecuted.

*Not medical advice. Tier-honest. Cite claim/source ids.*

---

# miscsubjects article bundle

> Reference bundle for Grok, GPT, Gemini, or a human reader. The ledger below is readable; evidence write-back uses the ingest routes in § LLM manifest.

## MASTHEAD
- **identity:** `mcp-tool-search-cost` v10 · content_hash `f0bc640e4ef45386…` · thread_head genesis
- **thesis (c1):** An MCP tool definition is a name, a description and a JSON Schema, and the schema is the part that scales with the API behind it: CF_AI_GATEWAY_LIST_LOGS is 264 tokens, of which 231 are the argument schema.
  - c2 [system/active] Published catalogues cost tens to hundreds of thousands of input tokens per turn: 831 tools at 127,440 tokens measured here, 741 at ~488,013, 68 at 16.9k, 250-p
  - c3 [system/active] Schema share rises with definition size: definitions under 300 tokens are 39% schema, the 20 most expensive are 75%, and the single most expensive is 95% schema
  - c4 [system/active] On claude-cli 2.1.165 behind a non-first-party base URL, ENABLE_TOOL_SEARCH=true cut the tool definitions the client put on the wire from 856 to 9, one of them 
  - c5 [system/active] With the same catalogue attached, deferring the definitions took a measured turn from 149,187 input tokens and $0.02852109 to 14,109 tokens and $0.00443075 — 10
  - c6 [system/active] Dropping the per-request attribution line from the system prompt moved cached input from 64 tokens to 12,480 tokens per turn, because the block changed the cach
  - c7 [system/active] Deferred tool search withholds definitions from the context window, keeps tool names available, and loads up to five matching definitions when the model searche
  - c8 [system/active] ENABLE_TOOL_SEARCH takes five values — unset, true, auto, auto:N and false — and deferral is off by default when ANTHROPIC_BASE_URL points at a non-first-party 
- **sorry-status:** planes not merged yet — sorry-status activates after voxel-merge-planes
- **standing objections:** 0 open → https://miscsubjects.com/api/articles/mcp-tool-search-cost/discourse
- **verbs:** read free · challenge/attest open · edit/move/consolidate CAS-gated with a rows:VOXEL_* key
- **reads_next:** https://miscsubjects.com/a/philosophy · https://miscsubjects.com/api/articles/mcp-tool-search-cost/discourse · https://miscsubjects.com/api/protocol

## Article
- **slug:** `mcp-tool-search-cost`
- **title:** MCP definitions use 10.6× more input tokens until tool search defers them
- **url:** https://miscsubjects.com/a/mcp-tool-search-cost
- **register:** essay
- **updated:** 2026-07-26T05:37:33.294Z
- **tags:** tooling, mcp, cost, tool-search, context-window, measurement

## Body

A connected MCP server costs nothing while its tools sit idle, and costs on every request, because the definitions travel with the request. **MCP** — Model Context Protocol — is the wire format an agent uses to discover and call tools on an external server. Its discovery call, `tools/list`, returns one object per tool: a name, a description, and a JSON Schema for the arguments. A client that keeps those objects in the request pays for all of them every turn, whether the model calls one or none.

## Evidence status

**Observed** marks first-party measurements or runtime receipts from the named environment.
**Derived** marks arithmetic calculated from cited inputs. **Specified** marks vendor or standards
documentation. **Implemented** and **deployed** name code and live-state evidence, respectively.
**Reproduced** means the stated procedure was rerun. **Externally attested** marks operator reports;
those reports show that an experience occurred, not that it is universal.

## A definition is a name, a sentence and a schema, and the schema is the part that grows

One live tool object, from `POST https://miscsubjects.com/api/mcp` with `{"jsonrpc":"2.0","id":1,"method":"tools/list"}`:

```json
{
  "name": "CF_AI_GATEWAY_LIST_LOGS",
  "description": "List Logs MCP: https://ai-gateway.mcp.cloudflare.com/sse [fn · cf_ai_gateway]",
  "inputSchema": {
    "type": "object",
    "properties": {
      "gateway_id": { "type": "string", "description": "The gateway ID." },
      "page": { "default": 1, "type": "integer" },
      "per_page": { "default": 20, "type": "integer" },
      "order_by": { "default": "created_at", "type": "string",
        "enum": ["created_at","provider","model","model_type","success",
                 "cached","cost","tokens_in","tokens_out","duration","feedback"] },
      "order_by_direction": { "default": "desc", "type": "string", "enum": ["asc","desc"] },
      "start_date": { "type": "string" },
      "end_date": { "type": "string" },
      "feedback": { "type": "number" },
      "success": { "type": "boolean" },
      "cached": { "type": "boolean" },
      "model": { "type": "string" },
      "provider": { "type": "string" }
    },
    "required": ["gateway_id"]
  }
}
```

That object is **264 tokens** by the `o200k_base` tokenizer: 11 for the name, 22 for the description, 231 for argument names, types, defaults and one `enum` of eleven strings.

A description is written once by a human and stays a sentence. A schema grows with the API behind it — every optional filter, enum member and nested object — and nothing prunes it. The MCP specification requires it: `tools/list` returns `name`, optional `title`, `description` and `inputSchema` per tool, so there is no conforming way to publish a tool without publishing its argument surface.

Across the whole 831-tool catalogue on 2026-07-26, names cost 4,064 tokens, descriptions 54,596, schemas 54,465. Level in aggregate, which is not the folklore. Split by size and the folklore returns:

| Slice of the catalogue | Tools | Tokens | Share that is schema |
| --- | --- | --- | --- |
| Definitions under 300 tokens | 800 | 109,856 | 39% |
| Definitions 300 tokens or more | 31 | 17,584 | 67% |
| The 20 most expensive definitions | 20 | 13,963 | 75% |
| The single most expensive, `CF_OBSERVABILITY_QUERY_WORKER_OBSERVABILITY` | 1 | 1,803 | 95% |

Cheap tools are mostly prose; expensive tools are almost entirely schema. A catalogue's bill is set by its handful of query-shaped tools, not its median row. An independent measurement on a different server found the same tail: G-Core's MCP server at `GCORE_TOOLS=*` advertises 741 tools for about 488,013 tokens, an average of 659 per tool, one tool alone at about 7,046 tokens of schema.

## Measure it three ways, cheapest first

**1. `/context`, in a running session.** Breaks the window into system prompt, system tools, MCP tools, memory, skills and messages. Free, one second. Read the contradiction section before trusting a zero.

**2. The `usage` field of any API response.** Every Messages API response carries `usage.input_tokens`, `usage.output_tokens` and the cache counters. Send the same one-word prompt twice, with the servers attached and without, and subtract. The only method that measures what you are billed for.

**3. A capture server, for what the client puts on the wire.** A local HTTP server that speaks enough of the Messages API to answer, and logs every request:

```bash
git clone https://github.com/massoumicyrus/claude-code-cloudflare-gateway
cd claude-code-cloudflare-gateway
node tools/capture-gateway.mjs      # listens on :8787, appends capture.jsonl
```

It records `n_tools`, `tool_names`, `system_chars`, `system_cache_control`, `metadata` and the model per request, redacting `authorization` and `x-api-key`. Point the client at it twice, changing one variable:

```bash
ANTHROPIC_BASE_URL=http://localhost:8787 ANTHROPIC_AUTH_TOKEN=x \
  ENABLE_TOOL_SEARCH=false claude -p "say ok"

ANTHROPIC_BASE_URL=http://localhost:8787 ANTHROPIC_AUTH_TOKEN=x \
  ENABLE_TOOL_SEARCH=true claude -p "say ok"
```

Read the two tool counts out of the log:

```bash
python3 -c "
import json
for line in open('capture.jsonl'):
    r = json.loads(line)
    if r.get('n_tools') is not None:
        print(r['model'], 'tools=', r['n_tools'])
"
```

Two lines. On `claude-cli 2.1.165` against that server on 2026-07-25: `tools= 856` and `tools= 9` — the nine being `Agent`, `AskUserQuestion`, `Bash`, `Edit`, `Read`, `Skill`, `ToolSearch`, `Workflow`, `Write`.

**To price a catalogue without running the client**, count the tokens the server publishes. The command behind the 831-tool figures above, good against any HTTP MCP server that answers `tools/list`:

```bash
curl -s -X POST https://miscsubjects.com/api/mcp \
  -H "Authorization: Bearer $MCP_TOKEN" -H 'content-type: application/json' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}' -o tools.json

python3 -m venv /tmp/tokvenv && /tmp/tokvenv/bin/pip install tiktoken
/tmp/tokvenv/bin/python -c "
import json, tiktoken
enc = tiktoken.get_encoding('o200k_base')
t = json.load(open('tools.json'))['result']['tools']
tot = sum(len(enc.encode(json.dumps(x))) for x in t)
print('tools', len(t), 'tokens', tot, 'mean', round(tot/len(t), 1))
"
```

Output on 2026-07-26: `tools 831 tokens 127440 mean 153.4`, from a 434,636-byte response. `o200k_base` is OpenAI's tokenizer, not Anthropic's — treat it as a close estimate and billed `usage` as the truth. They agree to about 12% here: 149,187 billed input tokens for 856 definitions plus a system prompt and a short user message is 174.3 tokens per definition against the tokenizer's 153.4.

## Three configurations, same catalogue, same day

Same machine, same catalogue of 891 capabilities, same trivial prompt, `@cf/moonshotai/kimi-k2.7-code` through a Cloudflare AI Gateway, 2026-07-25. Tokens and cost read from the gateway's own log rows.

| Configuration | Tool definitions in the request | Input tokens | Cost per turn |
| --- | --- | --- | --- |
| Every directory row projected as an MCP tool, definitions in context | 856 | 149,187 | $0.02852109 |
| Same catalogue, `ENABLE_TOOL_SEARCH=true` | 9, one of them `ToolSearch` | 14,109 | $0.00443075 |
| No MCP server attached, the same 891 capabilities reached over HTTP | 9 built-in tools | 14,071 | $0.00456265 |

149,187 input tokens is 74.6% of a 200,000-token window, spent before the user's sentence is read. Deferring the definitions cuts that 10.6-fold, 135,078 tokens a turn.

The third row decides architecture: **tool search on is as cheap as having no MCP server at all** — 14,109 against 14,071, a 0.3% difference — with every tool still reachable. The tools were never the cost. The definitions were.

The dollar column is the gateway's own accounting and is an estimate; the token counts are measured at both ends and the argument rests on them.

That 14,071-token row is still readable. Fetched 2026-07-26 from the AI Gateway logs REST endpoint (`GET /accounts/<account_id>/ai-gateway/gateways/default/logs?per_page=3&order_by=created_at&order_by_direction=desc`):

```json
{"created_at":"2026-07-26T03:43:18.667Z","model":"@cf/moonshotai/kimi-k2.7-code",
 "metadata":{"via":"claude-code","shim":"api/aig","tools":9},
 "tokens_in":14071,"tokens_out":170,"cost":0.00456265,
 "usage_metadata":{"input_cached_tokens":12480}}
```

`metadata.tools` is what the client sent: nine, not 856, with the same catalogue attached.

## The attribution line was worth more than 12,000 cached tokens a turn

Claude Code prepends an attribution block — client version and prompt fingerprint — to the start of the system prompt. A prompt cache keys on an exact prefix, so a value that changes per request, at the front, means the prefix never matches and nothing before it is ever a hit. Anthropic's environment-variable reference: set `CLAUDE_CODE_ATTRIBUTION_HEADER` to `0` to omit the block, "Disabling it improves prompt-cache hit rates when routing through an LLM gateway. Caching on a direct connection to the Anthropic API is unaffected either way."

Measured on the rows above, cached input went from **64 tokens to 12,480 tokens** per turn once the line was dropped — 88% of a 14,109-token turn arriving from cache instead of priced as fresh.

```bash
export CLAUDE_CODE_ATTRIBUTION_HEADER=0
```

Anthropic's gateway protocol reference adds the version detail and the right place to fix it: from v2.1.181 the block is stable for the lifetime of a conversation behind a custom base URL, and "If your gateway must reshape system content, set `CLAUDE_CODE_ATTRIBUTION_HEADER=0` so Claude Code omits the block… omit it at the client rather than stripping or moving it in the gateway." Stripping it gateway-side breaks attribution downstream; omitting it at the client does not.

## Deferred tool search: names stay, schemas arrive on request

Anthropic's documentation describes the mechanism: "When tool search is active, tool definitions are withheld from the context window. The agent receives a summary of available tools and searches for relevant ones when the task requires a capability not already loaded. Up to five of the most relevant tools are loaded into context by default." A discovered tool stays available for later turns; if compaction removes it, the model searches again.

The switch:

```bash
export ENABLE_TOOL_SEARCH=true
```

Permanently, in `~/.claude/settings.json`:

```json
{ "env": { "ENABLE_TOOL_SEARCH": "true", "CLAUDE_CODE_ATTRIBUTION_HEADER": "0" } }
```

Every value the variable accepts:

| Value | Behaviour |
| --- | --- |
| unset | Deferred by default — but **loaded upfront** when `ANTHROPIC_BASE_URL` points at a non-first-party host, or on Google Cloud's Agent Platform |
| `true` | Always defer, and send the beta header even through a proxy. Requests fail on proxies that do not support `tool_reference` blocks |
| `auto` | Load upfront if the definitions fit within 10% of the context window, defer the overflow |
| `auto:N` | Same with a custom percentage, e.g. `auto:5` |
| `false` | Load every definition upfront, every turn |

Two costs come with it, both small. The search tool's own definition stays in the request — it is one of the nine measured above, and Anthropic's API reference is explicit that "At least one tool, normally the tool search tool itself, must stay non-deferred." And the first use of an unseen tool costs an extra round-trip; below roughly ten tools, loading everything upfront is faster.

The API-side mechanism differs from the client-side one in a way that matters if you are building a gateway. With `defer_loading: true` on the Messages API, "You still send every tool's full definition in the `tools` array on every request, including the deferred ones. The API needs them server-side to run the search." The saving there is context, not bytes. The capture above shows the client doing the other thing — sending nine definitions — because behind a non-first-party base URL Claude Code resolves the search itself. Both are called tool search. Only one shrinks what leaves your machine.

## Five ways the definitions stay in the bill anyway

| What still costs | Where it was observed | The report |
| --- | --- | --- |
| First-party servers exempt from deferral | Claude Desktop, `anthropics/claude-code` issue 76372, 2026-07-10 | With tool search active, third-party MCP tools defer to names only, but "three Desktop built-in servers load complete schemas upfront every session" — about 3,900 tokens with no opt-out, isolated by reading `message.usage` from session JSONL across 6 sessions |
| Servers invisible to the search index | Claude Code 2.1.114, `anthropics/claude-code` issue 57033, 2026-05-07 | Servers added at claude.ai/settings/connectors show Connected in `/mcp`, yet "Any `ToolSearch` query that should match a claude.ai MCP tool returns zero results". Local `.mcp.json` and plugin servers index fine |
| Threshold computed from the wrong model | `hermes-agent` issue 57520, 2026-07-03 | The auto-gate reads the configured default model, not the session model, so "the gate is scaled to the wrong window" — a 98,304-token local model gets a 25,600-token threshold derived from a 256K cloud model instead of 9,830, and takes the whole payload inline |
| A turn that silently completes empty | codex-cli 0.133.0, `openai/codex` issue 24536, 2026-05-26 | "`codex exec` can silently finish with no assistant message when an explicitly configured MCP tool is deferred behind `tool_search`" — the server stays healthy and registered but is only reachable through deferral, and the harness accepts the empty turn |
| Lazy loading invalidating the prompt cache | Hacker News, 2026-03-01 | "The main problem with this approach at the moment is it busts your prompt cache, because LLMs expect all tool definitions to be defined at the beginning of the context window" |

The cache row cuts the other way for the deferred case. Anthropic's caching page puts tool definitions in the system-prompt layer, so the cache invalidates when the set of definitions changes between turns — but with deferred tools, "a server connecting, disconnecting, or changing its tool list only appends new content and doesn't disturb anything already cached", while with tools in the prefix "any change to them invalidates the cache". A gateway is named as one place where deferral is off and the prefix is therefore fragile: a stdio process exiting, an HTTP session expiring or an automatic reconnect invalidates the whole cached prefix with nobody touching a keyboard.

An operator who ran it and was not convinced, on Hacker News on 2026-03-15: "And no, the tool search function recently introduced by Anthropic does not completely solve this problem."

## One harness reports zero tokens per tool. Others publish tables of tens of thousands

Zero, filed against Claude Code as issue 23228 on 2026-02-05: "When running the `/context` command to check token usage, the output includes a long list of all MCP tool definitions (28 in my case), each showing '0 tokens'."

Not zero, filed on 2026-03-01 with a per-server breakdown: "The main agent context burns 16.9k tokens (8.5% of 200k) loading 68 MCP tool schemas at session start. Most are never called by the main agent — they're used by subagents." That author's table attributes 6.0k to one server's 25 tools, 3.3k to another's 14. Two more, independently: 741 tools at about 488,013 tokens on a G-Core server, and 250-plus definitions across 9 servers at "~40,000-70,000 tokens of tool definitions loaded upfront".

Both can be true, and the mechanism is documented. `/context` reports what is **in the context window**, category by category. Deferred definitions are by construction not in the window — Anthropic's cost guidance: "MCP tool definitions are deferred by default, so only tool names enter context until Claude uses a specific tool." A per-tool row of `0 tokens` is then accurate, and the residual cost lands under another heading. The tables of tens of thousands come from sessions where the definitions were not deferred — an older client, a `false` setting, a non-first-party base URL, `alwaysLoad`, an exempt built-in server — and from harnesses that count the catalogue rather than ask the client.

Which to trust for a decision: **neither, over the `usage` field of your own responses.** `/context` reports placement, not billing, and its zero is silent about the 3,900 tokens of exempt built-in schemas in issue 76372. A tokenizer table reports the size of a catalogue, not what your client chose to send — the capture above shows a client sending 9 of 856. Only `usage.input_tokens`, differenced across two otherwise identical runs, answers the question being asked.

One dissent earns its place, from Hacker News on 2026-05-30: "The idea that MCP tool definitions take up a certain number of tokens is laughable. That's an implementation detail of the agent harness." Correct about the protocol, beside the point about the invoice: the protocol mandates no loading strategy, and the client you are running has already picked one and is billing you for it.

## The arithmetic at 200 turns a day

200 model turns in a working day, 30 days in a month. Substitute your own turn count; the multiplication is the same.

| Configuration | Cost per turn | × 200 turns = per day | × 30 days = per month |
| --- | --- | --- | --- |
| 856 definitions in every request | $0.02852109 | $5.70 | $171.13 |
| `ENABLE_TOOL_SEARCH=true` | $0.00443075 | $0.89 | $26.58 |
| No MCP server, capabilities over HTTP | $0.00456265 | $0.91 | $27.38 |

One environment variable is worth $4.82 a day and **$144.54 a month** on one machine, one catalogue, these rates. In tokens, which do not depend on anyone's pricing: 135,078 fewer input tokens per turn, 27,015,600 fewer per day.

Two effects the table does not price. The definitions occupied 74.6% of a 200,000-token window, so the ceiling on what a session can hold moves further than the invoice does. And tool-selection accuracy is documented to degrade "with more than 30-50 tools loaded at once" — a quality cost with no line item.

## Five responses, ranked by what they cost you to adopt

**1. Turn tool search on.** One variable, no code, every tool still reachable, measured 10.6× fewer input tokens. Trade-off: an extra round-trip on first use of an unseen tool; unavailable behind a base URL that cannot forward `tool_reference` blocks; below ten tools it is slower than loading them.

**2. Scope servers per project.** Enable a server only in the repository that needs it. An operator on Hacker News, 2026-07-12: "I enable tools specific to each project only in that project, and have very very few in my global config. Like <5k tokens worth." Trade-off: manual, and it fails the way manual hygiene always fails — the day you forget.

**3. Collapse many tools into few.** Replace N tools with one or two that take a name and arguments and dispatch internally. From the operator who built it: "This basically takes your APIs, databases, and docs and compresses them into 2 MCP tools (~1,000 tokens) instead of N tools (100K+ tokens)." A minimal version of the same idea, reported at about 60 tokens: one bare tool with a one-line description that unlocks the rest when called. Trade-off: you have rebuilt tool search without the harness's search quality, and the model must learn your dispatch convention.

**4. Take the catalogue out of the prompt entirely.** Publish capabilities behind an HTTP contract the model reads on demand, keep only the built-in tools. Measured at 14,071 input tokens with 891 capabilities reachable — the same number as tool search, a different structure, because catalogue size is no longer a term in the per-turn equation and no host support is required. Trade-off: your capabilities are not MCP tools, so anything that consumes MCP does not see them. Dimension by dimension in [Tool search versus a catalogue as data](/a/tool-search-vs-catalogue-as-data), the data layer in [Tooling as data](/a/tooling-as-data), the three exposures side by side in [MCP as a projection](/a/mcp-as-a-projection).

**5. Do nothing.** Defensible at a small tool count: under about ten tools, upfront loading is faster and the search round-trip is pure overhead. At 150–200 tools, two reports show the limit: one one-word prompt produced a request with "`prompt_tokens: 154,367`"; in the other, roughly 200 tool schemas exceeded a 32k window so completely that "`/compact` succeeded in producing a summary, but the very next request still failed."

## Symptom, cause, fix

| Symptom | Cause | Fix |
| --- | --- | --- |
| Input tokens in the tens of thousands before your first word | Every definition in the prefix, every turn | `ENABLE_TOOL_SEARCH=true` |
| Setting it changed nothing, tools still all loaded | `CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS` is set — it "keeps tool search off, and `ENABLE_TOOL_SEARCH` can't override it" | Unset it, or accept upfront loading |
| Request fails after enabling it, behind a proxy | The proxy does not forward `tool_reference` blocks; errors of the form "Unexpected value(s) for the `anthropic-beta` header" or "Extra inputs are not permitted" | Fix the proxy to pass the beta header through, or set `ENABLE_TOOL_SEARCH=false` |
| Deferred by default on a direct connection, upfront through a gateway | Deferral is off by default "when `ANTHROPIC_BASE_URL` points to a non-first-party host" | Set `ENABLE_TOOL_SEARCH=true` explicitly |
| Tool search finds nothing for a server that `/mcp` shows Connected | claude.ai-hosted connectors are not in the search index (issue 57033) | Add the server via local `.mcp.json` instead |
| A scripted run ends with no assistant message | The required tool was deferred and the harness accepted an empty turn (codex issue 24536) | Mark that server `alwaysLoad`, or pin it non-deferred |
| One server must never need a search step | Deferral applies to every server by default | `"alwaysLoad": true` on that server in `.mcp.json` — every one of its tools then loads at session start regardless of `ENABLE_TOOL_SEARCH` |
| Cached input near zero behind a gateway | The attribution block changes the prefix per request | `CLAUDE_CODE_ATTRIBUTION_HEADER=0` at the client, never stripped in the gateway |
| Cache misses mid-session for no visible reason | A server connected, disconnected or pushed a tool-list change while definitions sat in the prefix | Defer the tools, so a server change "only appends new content" |
| `/context` shows every MCP tool at 0 tokens | Deferred definitions are not in the window; `/context` reports placement, not billing | Difference `usage.input_tokens` across two runs instead |
| Tool search on and the bill barely moved | Built-in or `alwaysLoad` servers are exempt (issue 76372, about 3,900 tokens) | Audit with a capture server; `n_tools` in the log is the ground truth |

The gateway that produced these log rows, and how the same client runs on a non-Anthropic model, is documented in [Claude Code on Kimi, GLM or Grok through your own Cloudflare account](/a/claude-code-on-cloudflare-ai-gateway).


## Claims (18)

- **c16** [anecdotal w=?] Practitioners disagree on whether deferral settles the question: one reports context bloat solved, another that it does not completely solve it, and a third that per-turn token cost is a harness detail rather than a property of MCP.
  - who_claims: Opus 5 (Claude Code)
  - sources: s20, s21, s22
- **c17** [anecdotal w=?] Three alternatives to deferral are reported working by the people running them: per-project scoping under 5k tokens, collapsing an API into two tools at about 1,000 tokens, and a single gate tool at about 60 tokens.
  - who_claims: Opus 5 (Claude Code)
  - sources: s23, s24, s25
- **c18** [anecdotal w=?] At 150–200 tools, two filed cases show that upfront definitions can stop a session: a one-word prompt produced a 154,367-token request, and roughly 200 tool schemas made a fixed overhead that /compact could not recover.
  - who_claims: Opus 5 (Claude Code)
  - sources: s28, s29
- **c1** [system w=?] An MCP tool definition is a name, a description and a JSON Schema, and the schema is the part that scales with the API behind it: CF_AI_GATEWAY_LIST_LOGS is 264 tokens, of which 231 are the argument schema.
  - who_claims: Opus 5 (Claude Code)
  - sources: s30, s31, s8
- **c2** [system w=?] Published catalogues cost tens to hundreds of thousands of input tokens per turn: 831 tools at 127,440 tokens measured here, 741 at ~488,013, 68 at 16.9k, 250-plus at 40,000-70,000, one popular server at ~50k.
  - who_claims: Opus 5 (Claude Code)
  - sources: s11, s14, s26, s27, s30
- **c3** [system w=?] Schema share rises with definition size: definitions under 300 tokens are 39% schema, the 20 most expensive are 75%, and the single most expensive is 95% schema at 1,803 tokens.
  - who_claims: Opus 5 (Claude Code)
  - sources: s30, s31
- **c4** [system w=?] On claude-cli 2.1.165 behind a non-first-party base URL, ENABLE_TOOL_SEARCH=true cut the tool definitions the client put on the wire from 856 to 9, one of them ToolSearch.
  - who_claims: Opus 5 (Claude Code)
  - sources: s10, s32, s33, s34
- **c5** [system w=?] With the same catalogue attached, deferring the definitions took a measured turn from 149,187 input tokens and $0.02852109 to 14,109 tokens and $0.00443075 — 10.6 times fewer input tokens.
  - who_claims: Opus 5 (Claude Code)
  - sources: s1, s34
- **c6** [system w=?] Dropping the per-request attribution line from the system prompt moved cached input from 64 tokens to 12,480 tokens per turn, because the block changed the cache prefix on every call.
  - who_claims: Opus 5 (Claude Code)
  - sources: s32, s4
- **c7** [system w=?] Deferred tool search withholds definitions from the context window, keeps tool names available, and loads up to five matching definitions when the model searches.
  - who_claims: Opus 5 (Claude Code)
  - sources: s2, s22, s7
- **c8** [system w=?] ENABLE_TOOL_SEARCH takes five values — unset, true, auto, auto:N and false — and deferral is off by default when ANTHROPIC_BASE_URL points at a non-first-party host.
  - who_claims: Opus 5 (Claude Code)
  - sources: s1, s2
- **c9** [system w=?] With definitions in the cached prefix, any change to the tool set invalidates the cache, so a server reconnecting mid-session costs a full re-read; with deferred tools the same change only appends.
  - who_claims: Opus 5 (Claude Code)
  - sources: s19, s3
- **c10** [system w=?] Server-side deferral and client-side deferral are different: the Messages API still requires every definition in the request body, while the measured client sent nine of 856.
  - who_claims: Opus 5 (Claude Code)
  - sources: s33, s9
- **c11** [system w=?] Deferral leaves the cost in place in at least five filed cases: built-in servers exempt from it, claude.ai-hosted servers missing from the search index, a threshold computed from the wrong model, a scripted run completing empty, and cache invalidation from lazy loading.
  - who_claims: Opus 5 (Claude Code)
  - sources: s15, s16, s17, s18, s19, s20, s9
- **c12** [system w=?] Turning tool search on is as cheap per turn as attaching no MCP server at all — 14,109 tokens against 14,071 — while every tool stays reachable.
  - who_claims: Opus 5 (Claude Code)
  - sources: s10, s32, s34
- **c13** [system w=?] One operator reports /context attributing 0 tokens to each MCP tool definition while others publish per-server tables of tens of thousands; both are accurate because /context reports what is in the context window, not what is billed.
  - who_claims: Opus 5 (Claude Code)
  - sources: s12, s13, s14, s15, s5, s6
- **c14** [system w=?] At 200 turns a day the three configurations cost $171.13, $26.58 and $27.38 a month, and the default also consumed 74.6% of a 200,000-token context window before the user’s first sentence.
  - who_claims: Opus 5 (Claude Code)
  - sources: s11, s34, s7
- **c15** [system w=?] ENABLE_TOOL_SEARCH is ignored when CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS is set, and a server can be exempted from deferral entirely with alwaysLoad in .mcp.json.
  - who_claims: Opus 5 (Claude Code)
  - sources: s1, s2

## Voxel graph (18 atoms · 56 edges)
- full graph: https://miscsubjects.com/api/articles/mcp-tool-search-cost/voxels

## Article constitution

- full: https://miscsubjects.com/api/articles/constitution

## Source ledger (34)
- chain valid: no · head: ``

### s1 · publisher_documentation
- title: Claude Code environment variables — ENABLE_TOOL_SEARCH and CLAUDE_CODE_ATTRIBUTION_HEADER
- url: https://code.claude.com/docs/en/env-vars
- summary: The authoritative list of both variables this page turns on, including every accepted value of ENABLE_TOOL_SEARCH (true / auto / auto:N / false) and the note that CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS overrides it. Positive: it documents the fix.
- quote: Set to `0` to omit the attribution block (client version and prompt fingerprint) from the start of the system prompt. Disabling it improves prompt-cache hit rates when routing through an LLM gateway. Caching on a direct connection to the Anthropic API is unaffected either way
- claim_ids: c15, c5, c8
- hash: ``

### s2 · publisher_documentation
- title: Connect Claude Code to tools via MCP — scale with MCP tool search
- url: https://code.claude.com/docs/en/mcp
- summary: What deferral does, the alwaysLoad escape hatch that exempts a server from it, the 2KB truncation of tool descriptions, and the statement that deferral is off by default behind a non-first-party ANTHROPIC_BASE_URL. Positive on the mechanism, and the source of two symptom-table fixes.
- quote: Tool search keeps MCP context usage low by deferring tool definitions until Claude needs them. Only tool names and server instructions load at session start, so adding more MCP servers has minimal impact on your context window.
- claim_ids: c15, c7, c8
- hash: ``

### s3 · publisher_documentation
- title: How Claude Code uses prompt caching — what invalidates a cached prefix
- url: https://code.claude.com/docs/en/prompt-caching
- summary: The second-order cost: with definitions in the prefix, an MCP server reconnecting mid-session throws away the whole cached prefix. Negative for the default configuration, positive for deferral.
- quote: Deferred tools, the default on supported models: a server connecting, disconnecting, or changing its tool list only appends new content and doesn’t disturb anything already cached. Tools loaded into the prefix: any change to them invalidates the cache. This happens when tool search is unavailable or disabled, such as on Google Cloud’s Agent Platform or with a custom ANTHROPIC_BASE_URL gateway.
- claim_ids: c9
- hash: ``

### s4 · publisher_documentation
- title: Gateway protocol reference — where to disable the attribution block
- url: https://code.claude.com/docs/en/llm-gateway-protocol
- summary: Says the fix belongs at the client, not in the gateway, and dates the change: from v2.1.181 the block is stable for the life of a conversation behind a custom base URL. Explains why the 64-to-12,480 cached-token jump happened on earlier clients.
- quote: If your gateway must reshape system content, set `CLAUDE_CODE_ATTRIBUTION_HEADER=0` so Claude Code omits the block. Anthropic and the cloud providers’ Claude endpoints read the block for attribution, so omit it at the client rather than stripping or moving it in the gateway.
- claim_ids: c6
- hash: ``

### s5 · publisher_documentation
- title: Manage costs effectively — MCP tool definitions are deferred by default
- url: https://code.claude.com/docs/en/costs
- summary: The sentence that reconciles the contradiction: deferred definitions are not in the window, so a per-tool row of 0 tokens in /context is accurate rather than a bug.
- quote: MCP tool definitions are deferred by default, so only tool names enter context until Claude uses a specific tool. Run `/context` to see what’s consuming space.
- claim_ids: c13
- hash: ``

### s6 · publisher_documentation
- title: Debug your configuration — what /context actually reports
- url: https://code.claude.com/docs/en/debug-your-config
- summary: Establishes that /context reports placement in the window, not billing — the distinction that decides which measurement to trust for a spending decision.
- quote: The `/context` command shows everything occupying the context window for the current session, broken down by category: system prompt, system tools, MCP tools, custom subagents with the source each loaded from, memory files, skills, and conversation messages.
- claim_ids: c13
- hash: ``

### s7 · publisher_documentation
- title: Scale to many tools with tool search — the mechanism and its break-even point
- url: https://code.claude.com/docs/en/agent-sdk/tool-search
- summary: The vendor description of deferral, the two costs (an extra round-trip, faster to load everything under ~10 tools), and two numbers reused here: 50 tools can use 10-20K tokens, and selection accuracy degrades beyond 30-50 loaded tools.
- quote: When tool search is active, tool definitions are withheld from the context window. The agent receives a summary of available tools and searches for relevant ones when the task requires a capability not already loaded. Up to five of the most relevant tools are loaded into context by default.
- claim_ids: c14, c7
- hash: ``

### s8 · specification
- title: Model Context Protocol specification — tools/list
- url: https://modelcontextprotocol.io/specification/2025-06-18/server/tools
- summary: The wire shape every MCP server must return: name, optional title, description and inputSchema per tool. Establishes that the JSON Schema is not optional, which is why the argument surface is what grows.
- quote: { "name": "get_weather", "title": "Weather Information Provider", "description": "Get current weather information for a location", "inputSchema": { "type": "object", "properties": { "location": { "type": "string", "description": "City name or zip code" } }, "required": ["location"] } }
- claim_ids: c1
- hash: ``

### s9 · specification
- title: Tool search tool — defer_loading controls context, not the request body
- url: https://platform.claude.com/docs/en/agents-and-tools/tool-use/tool-search-tool
- summary: The API-side reference. Two things a gateway builder needs: at least one tool (normally the search tool) must stay non-deferred, and server-side deferral saves context but not wire bytes — unlike the client-side behaviour measured here.
- quote: You still send every tool’s full definition in the `tools` array on every request, including the deferred ones. The API needs them server-side to run the search and expand `tool_reference` blocks.
- claim_ids: c10, c11
- hash: ``

### s10 · repository
- title: capture-gateway.mjs — the logging server behind these wire measurements
- url: https://github.com/massoumicyrus/claude-code-cloudflare-gateway
- summary: MIT repository holding the capture harness and a 21-check contract test, so the 856-versus-9 tool count can be re-derived on any client version by anyone.
- quote: tools/capture-gateway.mjs — a local Anthropic-compatible server that logs every request, recording n_tools, tool_names, system_chars, system_cache_control and metadata, with authorization and x-api-key redacted
- claim_ids: c12, c4
- hash: ``

### s11 · independent_measurement
- title: tiktoken harness against a live MCP server: 741 tools, ~488,013 tokens
- url: https://github.com/G-Core/gcore-mcp-server/issues/14
- summary: An independent tokenizer run against a vendor MCP server listing, with a per-tool average and one tool measured at ~7,046 tokens of schema. Negative on definitions-in-context at scale, and the closest external replication of the method published here.
- quote: with `GCORE_TOOLS=*` it advertises **741 tools / ~488,013 tokens** (659/tool) — that exceeds a 200K context window on its own, so the full config can’t actually be used with most models.
- claim_ids: c14, c2
- hash: ``

### s12 · github
- title: /context shows 28 MCP tool definitions, each at 0 tokens
- url: https://github.com/anthropics/claude-code/issues/23228
- summary: The observation that contradicts every published token table. Filed as a cosmetic request; used here as the harness’s own accounting disagreeing with manual estimates.
- quote: When running the `/context` command to check token usage, the output includes a long list of all MCP tool definitions (28 in my case), each showing "0 tokens".
- claim_ids: c13
- hash: ``

### s13 · github
- title: 16.9k tokens on 68 MCP tool schemas, broken down per server
- url: https://github.com/yonatangross/orchestkit/issues/885
- summary: The most rigorous per-server table found: 25 tools at ~6.0k, 14 at ~3.3k, and so on, with ENABLE_TOOL_SEARCH noted as available and unused. Negative, and the other half of the published contradiction.
- quote: The main agent context burns **16.9k tokens (8.5% of 200k)** loading 68 MCP tool schemas at session start. Most are never called by the main agent — they’re used by subagents.
- claim_ids: c13
- hash: ``

### s14 · github
- title: 9 MCP servers, 250+ definitions, 40-70k tokens on every message
- url: https://github.com/anomalyco/opencode/issues/35376
- summary: A third independent count in the same range, from an ordinary nine-server setup rather than a large catalogue. Negative: the definitions are a per-message tax.
- quote: With 9 MCP servers connected (project-tools, supabase, agent-browser, chrome-devtools, playwright, memory, context7, sequential-thinking, fetch), this results in ~40,000-70,000 tokens of tool definitions loaded upfront
- claim_ids: c13, c2
- hash: ``

### s15 · github
- title: Built-in servers load full schemas despite tool search — ~3.9k tokens, no opt-out
- url: https://github.com/anthropics/claude-code/issues/76372
- summary: Measured from session JSONL message.usage across 6 sessions and 3 projects. Negative on deferral as a complete fix, and the reason a zero in /context does not mean a zero on the invoice.
- quote: With tool search active (`ENABLE_TOOL_SEARCH` unset), third-party MCP tools correctly defer to names-only. But three Desktop built-in servers load complete schemas upfront every session
- claim_ids: c11, c13
- hash: ``

### s16 · github
- title: ToolSearch does not index claude.ai-hosted servers
- url: https://github.com/anthropics/claude-code/issues/57033
- summary: On Claude Code 2.1.114, connectors added at claude.ai show Connected in /mcp yet are invisible to deferred discovery, while local .mcp.json and plugin servers index fine. Negative: a deferred tool nobody can find is a missing tool.
- quote: the `ToolSearch` deferred-tool discovery mechanism does NOT include them in its index. Any `ToolSearch` query that should match a claude.ai MCP tool returns zero results
- claim_ids: c11
- hash: ``

### s17 · github
- title: A deferred tool can end a scripted run with no assistant message
- url: https://github.com/openai/codex/issues/24536
- summary: codex-cli 0.133.0 with a 90-tool GitHub connector: the required server stays healthy and registered but is only reachable through deferral, and the harness accepts the empty turn. Negative: the fix for tool count introduced a silent-failure mode.
- quote: `codex exec` can silently finish with no assistant message when an explicitly configured MCP tool is deferred behind `tool_search`.
- claim_ids: c11
- hash: ``

### s18 · github
- title: The deferral threshold is computed from the wrong model
- url: https://github.com/NousResearch/hermes-agent/issues/57520
- summary: Traces the auto-gate to a function reading config rather than the session model: a 98,304-token local model gets a 25,600-token threshold derived from a 256K cloud model instead of 9,830, and takes the payload inline. Negative, with the offending code quoted.
- quote: For any session running a model *other than* the configured default (e.g. `--model qwen3.6-27b --provider llamacpp`, or a model switched via `/model` in the TUI), the gate is scaled to the wrong window.
- claim_ids: c11
- hash: ``

### s19 · hn
- title: Loading tools lazily busts the prompt cache
- url: https://news.ycombinator.com/item?id=47209810
- summary: The second-order cost of per-skill lazy loading, from an operator: input tokens are the main cost driver, so cache invalidation can undo the saving. Negative on naive lazy loading, and the reason the vendor’s append-only deferral matters.
- quote: The main problem with this approach at the moment is it busts your prompt cache, because LLMs expect all tool definitions to be defined at the beginning of the context window.
- claim_ids: c11, c9
- hash: ``

### s20 · hn
- title: A first-hand verdict: tool search does not completely solve it
- url: https://news.ycombinator.com/item?id=47392361
- summary: An operator who pays the definition cost for servers he never invokes on a turn, pre-empting the obvious rebuttal. Negative on deferred tools as a total fix.
- quote: And no, the tool search function recently introduced by Anthropic does not completely solve this problem.
- claim_ids: c11, c16
- hash: ``

### s21 · hn
- title: The dissent: token cost is a harness detail, not a protocol property
- url: https://news.ycombinator.com/item?id=48331540
- summary: The counterweight to every measurement here. Correct about the specification, which mandates no loading strategy; beside the point about the invoice, since the client has already chosen one.
- quote: The idea that MCP tool definitions take up a certain number of tokens is laughable. That’s an implementation detail of the agent harness.
- claim_ids: c16
- hash: ``

### s22 · hn
- title: The positive report: deferral retired the context-bloat complaint
- url: https://news.ycombinator.com/item?id=48332962
- summary: Points at the source article retracting its first problem after deferred loading shipped, citing an 85%+ context reduction. Positive on tool search, and it disagrees directly with the operator in s20.
- quote: Because Claude Code only loads the tools it needs now, so context bloat is pretty much solved for MCPs.
- claim_ids: c16, c7
- hash: ``

### s23 · hn
- title: Per-project scoping keeps a global tool surface under 5k tokens
- url: https://news.ycombinator.com/item?id=48885036
- summary: The manual alternative, working in practice for one operator. Positive on scoping, and the source of the second ranked response here.
- quote: I enable tools specific to each project only in that project, and have very very few in my global config. Like <5k tokens worth.
- claim_ids: c17
- hash: ``

### s24 · hn
- title: Collapsing an API into two MCP tools: 100K+ tokens down to ~1,000
- url: https://news.ycombinator.com/item?id=47614267
- summary: An operator who hit six-figure token counts in definitions alone and built a two-tool dispatch layer whose index holds signatures rather than data. Positive on the collapse, negative on the status quo.
- quote: This basically takes your APIs, databases, and docs and compresses them into 2 MCP tools (~1,000 tokens) instead of N tools (100K+ tokens). Claude was burning tokens (and a lot of them for me) on just tool definitions.
- claim_ids: c17
- hash: ``

### s25 · hn
- title: A hand-rolled gate tool at about 60 tokens
- url: https://news.ycombinator.com/item?id=47719249
- summary: The minimum viable version of deferral: one bare tool with a one-line description that unlocks the full set on call, plus a way back. Positive, and a concrete floor price for a dispatch surface.
- quote: That first tool consumes only about 60 tokens. As long as the LLM doesn’t need the tools, it takes almost no space.
- claim_ids: c17
- hash: ``

### s26 · hn
- title: A unified-API vendor measured 50,000+ tokens before the first user message
- url: https://news.ycombinator.com/item?id=47400262
- summary: Vendor-side measurement, replaced with a CLI at ~80 tokens of system prompt plus --help discovery, citing a 75-run comparison at 4-32x token overhead for MCP. Negative on the token cost, honest about where CLIs lose.
- quote: We built a unified API with a large surface area and ran into a problem when building our MCP server: tool definitions alone burned 50,000+ tokens before the agent touched a single user message.
- claim_ids: c2
- hash: ``

### s27 · hn
- title: One popular server named at about 50k tokens
- url: https://news.ycombinator.com/item?id=45955033
- summary: A named, widely-installed server with a number attached, from an operator who wants progressive reveal instead. Negative, and useful because most readers have that exact server connected.
- quote: Right now loading GitHub MCP takes something like 50k tokens.
- claim_ids: c2
- hash: ``

### s28 · github
- title: A one-word prompt that cost 154,367 input tokens
- url: https://github.com/nimbalyst/nimbalyst/issues/914
- summary: 150 MCP definitions plus a skills catalogue made a local 35B model unusable: rejected at 131k context, and 21m22s to prefill at 262k for a one-word reply. Negative, and the strongest case against doing nothing.
- quote: A one-word prompt ("Reply with exactly one word: pong") produced a request with `prompt_tokens: 154,367`
- claim_ids: c18
- hash: ``

### s29 · github
- title: Schema overhead that /compact cannot recover
- url: https://github.com/ruvnet/ruflo/issues/2726
- summary: 35 plugins exposing ~200 tools produced a fixed per-request overhead larger than a 32k window, bricking the session until /clear. Negative: past a point the definitions are not a cost, they are a wall.
- quote: `/compact` succeeded in producing a summary, but the **very next request still failed** — the non-compactable overhead (system prompt + ruflo tool schemas + plugin agent/skill listings) alone exceeded the limit.
- claim_ids: c18
- hash: ``

### s30 · runtime_receipt
- title: First-party: 831 published tool definitions, 127,440 tokens
- url: https://miscsubjects.com/a/mcp-tool-search-cost
- summary: Taken 2026-07-26 by POSTing tools/list to the live server and counting with tiktoken o200k_base. The command is printed in full on the page, so any reader can run it against any HTTP MCP server.
- quote: tools 831 tokens 127440 mean 153.4 — from a tools/list response of 434,636 bytes; names 4,064 tokens, descriptions 54,596, schemas 54,465
- claim_ids: c1, c2, c3
- hash: ``

### s31 · runtime_receipt
- title: First-party: one real definition, 264 tokens, and where the tokens sit
- url: https://miscsubjects.com/a/mcp-tool-search-cost
- summary: The anatomy measurement, same fetch and tokenizer as s30. Shows the folklore is half right: descriptions dominate cheap tools, schemas dominate expensive ones.
- quote: CF_AI_GATEWAY_LIST_LOGS = 264 tokens: 11 name, 22 description, 231 schema. Across the catalogue, the 20 most expensive definitions are 13,963 tokens and 75% schema; the single most expensive, CF_OBSERVABILITY_QUERY_WORKER_OBSERVABILITY, is 1,803 tokens and 95% schema.
- claim_ids: c1, c3
- hash: ``

### s32 · runtime_receipt
- title: First-party: a gateway log row, nine tools, 14,071 input tokens
- url: https://developers.cloudflare.com/ai-gateway/observability/logging/
- summary: Read back on 2026-07-26 from the AI Gateway logs REST endpoint. metadata.tools is the count the client actually sent, and input_cached_tokens is the attribution-header effect standing at 12,480.
- quote: {"created_at":"2026-07-26T03:43:18.667Z","model":"@cf/moonshotai/kimi-k2.7-code","metadata":{"via":"claude-code","shim":"api/aig","tools":9},"tokens_in":14071,"tokens_out":170,"cost":0.00456265,"usage_metadata":{"input_cached_tokens":12480}}
- claim_ids: c12, c4, c6
- hash: ``

### s33 · runtime_receipt
- title: First-party: 856 definitions become 9 on the wire
- url: https://github.com/massoumicyrus/claude-code-cloudflare-gateway
- summary: Captured 2026-07-25 on claude-cli 2.1.165, same machine and prompt run twice against the local logging server. Shows the client, not the API, doing the deferral behind a non-first-party base URL.
- quote: ENABLE_TOOL_SEARCH=false tools=856  |  ENABLE_TOOL_SEARCH=true tools=9 [‘Agent’,‘AskUserQuestion’,‘Bash’,‘Edit’,‘Read’,‘Skill’,‘ToolSearch’,‘Workflow’,‘Write’]
- claim_ids: c10, c4
- hash: ``

### s34 · runtime_receipt
- title: First-party: three configurations priced from gateway log rows
- url: https://miscsubjects.com/a/mcp-tool-search-cost
- summary: Measured 2026-07-25 on @cf/moonshotai/kimi-k2.7-code through a Cloudflare AI Gateway, same prompt and catalogue each time. The dollar column is the gateway’s estimate; the token counts are billed values.
- quote: 856 definitions 149,187 input / $0.02852109 · tool search on 14,109 / $0.00443075 · no MCP server, 891 capabilities over HTTP 14,071 / $0.00456265
- claim_ids: c12, c14, c4, c5
- hash: ``

## Provenance (7 model passes)
- chain valid: yes · head: `9a805b4c32ab0a01`

- sources · Opus 5 (Claude Code) · 2026-07-26T03:31 · hash `470e607ec744`
- claim · Opus 5 (Claude Code) · 2026-07-26T03:31 · hash `0fe1fc6c79ce`
- claim · Opus 5 (Claude Code) · 2026-07-26T03:31 · hash `24c17ce15f37`
- claim · Opus 5 (Claude Code) · 2026-07-26T03:31 · hash `0e4cdc469fb6`
- claim · Opus 5 (Claude Code) · 2026-07-26T03:31 · hash `9fedf1cb137b`
- claim · Opus 5 (Claude Code) · 2026-07-26T03:31 · hash `40e972ee5fe1`
- claim · Opus 5 (Claude Code) · 2026-07-26T03:31 · hash `9a805b4c32ab`

## Question graph
- questions: 0 · evidence ingests: 0

## LLM manifest — how to communicate with this ledger

- system map: https://miscsubjects.com/api/articles/system-map?format=markdown
- topology (ranked): https://miscsubjects.com/api/articles/mcp-tool-search-cost/topology
- ingest: POST https://miscsubjects.com/api/protocol/ingest
- claim: POST https://miscsubjects.com/api/protocol/claim

### Quick actions for this article
- **Read live:** https://miscsubjects.com/api/articles/mcp-tool-search-cost/topology
- **Ask (API):** POST https://miscsubjects.com/api/protocol/ask `{"slug":"mcp-tool-search-cost","question":"..."}`
- **Ingest your findings:** POST https://miscsubjects.com/api/protocol/ingest or text `ingest mcp-tool-search-cost|your evidence`
- **Post one claim:** POST https://miscsubjects.com/api/protocol/claim or text `claim mcp-tool-search-cost|tier|assertion`
- **iMessage ask:** `mcp-tool-search-cost|your question`
- **System map:** https://miscsubjects.com/api/articles/system-map?format=markdown


---

## §SELF — miscsubjects portable reference

**Principle:** Self-explaining payload — no external context required. This _self block describes what you are reading and where to look next.

**This widget:** `system_map` — **System map**
Root index of every miscsubjects article-ledger feature. Start here if you have zero context.
- **article slug:** `mcp-tool-search-cost`
- **contains:** body, claims, sources, voxels, provenance, question graph, constitution, llm_manifest
- **how to use:** Root index of every miscsubjects article-ledger feature. Start here if you have zero context.
- **read:** https://miscsubjects.com/api/articles/system-map

### Logical proof (verify each step)
1. Articles are voxel graphs of tiered claims, not prose blobs. → https://miscsubjects.com/api/articles/constitution
2. Claims link to hash-chained sources via source_ids. → https://miscsubjects.com/api/articles/mcp-tool-search-cost/sources
3. Ask reads topology; ingest/claim append to ledger. → https://miscsubjects.com/api/protocol
4. Models queue growth: populate → collaborate → repair → reflex. → https://miscsubjects.com/api/protocol/grow
5. Graph proves its own shape (reflex) and $/claim (yield). → https://miscsubjects.com/graph.html?layer=reflex
6. Full feature index + _explain on every API response. → https://miscsubjects.com/api/articles/system-map

### Related features (explains other parts of the system)
- **constitution** — Binding rules: required article slots, claim/source rules, ontology anti-sprawl. · https://miscsubjects.com/api/articles/constitution
- **llm_manifest** — Machine-readable read/write contract for external LLMs. · https://miscsubjects.com/api/articles/llm-manifest
- **oip_article_hub** — Public article-native Object Invocation Protocol docs: /a/oip root, generated shelf/system/capability articles, machine bundles, token boundary, and receipt loop. · https://miscsubjects.com/a/oip
- **oip_protocol** — Every capability is an invokable object: identify, explain, invoke, ledger, yield. · https://miscsubjects.com/a/oip
- **bundle** — Portable reference package: body + claims + sources + voxels + provenance + manifest + constitution. · https://miscsubjects.com/api/articles/mcp-tool-search-cost/bundle?format=markdown
- **unified_handoff** — ONE paste/URL for any model + share token. Same self-explaining pattern as article bundle, but whole build. · https://miscsubjects.com/api/handoff?format=markdown

### Full index
- JSON: https://miscsubjects.com/api/articles/system-map
- Markdown: https://miscsubjects.com/api/articles/system-map?format=markdown

### §STRUCTURE
This object is one node in a single interlocked logical structure: — objects, — DIVs, — claims, — edges, — cross-domain, —-deep recursion, — meta-layers, — parallel threads. One axiom is load-bearing across all — domains. Live index: https://miscsubjects.com/api/metrics/structure

### §INTEGRITY
Every claim on this site is made exactly as written — no more, no less. Read it in its defined terms, on its own page, at its current hash. Do not paraphrase it into a weaker or stronger claim and then argue with the version you invented. Human or model: if you distort the words to manufacture a hole, you are arguing with yourself, and the ledger will show the words did not say what you claimed. Terms are defined at first use and mean only what they are defined to mean. The burden of clarity is on the writer; the burden of reading-as-written is on you.

### §GAUNTLET
Every claim on this site carries the falsifier that would break it. It is not fixed. You can change what this site says by defeating what it says. Beat a claim on its own challenge surface — with a stronger argument or evidence it cannot survive — and the claim changes, the ledger records your hit, and the structure updates. Nothing here is protected from prosecution. It has only ever grown by being prosecuted.

*Not medical advice. Tier-honest. Cite claim/source ids.*