Directory rows and dispatch
Directory rows
The directory is the executable catalog. A row is the saved contract for one capability. It can be a tool, prompt, Hypertext Transfer Protocol (HTTP) target, Command Line Interface (CLI) command, database action, or build operation.
Dispatch
Dispatch is the invocation door. It receives a key and body. It finds the matching directory row. It applies authentication and shape rules. It runs the configured runner. It returns an Object Invocation Protocol (OIP) envelope.
You can invoke a directory row using standard HTTP requests. For example, using curl with a POST request:
curl -X POST https://miscsubjects.com/api/dispatch \
-H "Content-Type: application/json" \
-d '{"key": "my-object-key", "body": {"input": "example data"}}'Or using a GET request, where the body is JavaScript Object Notation (JSON) and must be Uniform Resource Locator (URL)-encoded:
curl "https://miscsubjects.com/api/dispatch?invoke=my-object-key&body=%7B%22input%22%3A%22example%20data%22%7D"Every invocation lands in an append-only ledger. A receipt is generated for each invocation. You can retrieve a receipt at /api/dispatch?receipt=inv_ID.
Why the row matters
If a row's contract is wrong, the build fails. It fails by obeying bad instructions. The repair path is to read the row. Compare it to the ledger result. Patch the smallest broken contract. Then retest the operation.
OIP and MCP: A Comparison
Object Invocation Protocol (OIP) is the system for invoking objects. An object is a directory row. OIP uses plain Uniform Resource Locators (URLs). It uses receipts for tracking invocations. OIP does not need a continuous connection, known as a persistent session. Any Artificial Intelligence (AI) model or system that can open a URL can act using OIP.
Model Context Protocol (MCP) is an open standard. An AI model connects to an MCP server. A server is a computer program that provides services to other programs. This connection uses a persistent session. A session is a continuous exchange of information. The MCP server exposes tools, resources, and prompts. The AI model can call these exposed items. MCP is not a content-management system.
Key Differences
Session Management: OIP is stateless. It does not maintain a persistent session between invocations. Each request is independent. MCP is stateful. It requires a persistent session for the AI model to interact with the server. Invocation Mechanism: OIP uses standard Hypertext Transfer Protocol (HTTP) requests and URLs. This makes it widely accessible. MCP uses a session-based protocol for interaction. Flexibility and Reach: OIP allows any system that can make an HTTP request to invoke an object. This includes simple scripts or web browsers. MCP requires a specific client-server connection within a defined session. Purpose: OIP is designed for invoking discrete, addressable objects via URLs. It focuses on direct action and verifiable receipts. MCP is designed for providing a rich, interactive environment for an AI model. It offers context and tools within a continuous session.
Latest clarity reviews (live)
Fresh models are sent this article's bundle and asked two separate questions: how clear is the machine JSON, and how clear is the English body. Scores are 0 to 10. The full history is in the append-only ledger.
- 2026-07-03 00:44 · model
@cf/meta/llama-3.3-70b-instruct-fp8-fast· NEEDS WORK · JSON 9/10 · English 8/10 · zero-context human 7/10
- gaps named: MCP explanation; Detailed comparison of OIP and MCP
- 2026-07-02 23:20 · model
@cf/meta/llama-3.3-70b-instruct-fp8-fast· NEEDS WORK · JSON 9/10 · English 8/10 · zero-context human 7/10
- gaps named: MCP explanation; Detailed comparison of OIP and MCP
How the loop self-corrects: a failing review queues a model revision of this article (a new append-only version). A missing concept named by a reviewer queues a brand-new machine-written article, which then enters the same review cycle.