M · MCP Server

MCP server

Read and search the public framework from an MCP-aware client through a deterministic, read-only endpoint.

The framework exposes a Model Context Protocol endpoint at https://decision-grade.ai/api/mcp.

It publishes the same page content, search index, and full-text bundle available on the site. The server does not generate an answer or call an external model. Your chosen client does the reading and reasoning.

Endpoint: https://decision-grade.ai/api/mcp
Transport: HTTP and JSON-RPC 2.0
Tools: list_pages, get_page, search, get_full_framework
Resources: llms.txt, llms-full.txt

Why the endpoint exists

The public argument should be inspectable in the reader's own tools.

The reader chooses the model

Connect from any compatible client. The framework does not choose the AI that interprets it.

The server returns source text

Tool calls return published Markdown or search results. There is no hidden synthesis layer between the page and the client.

Tools

list_pages

Returns all framework pages with their slug, title, number, and description.

Arguments: none.

get_page

Returns one page from the full-text bundle.

Argument: slug. One of introduction, the-frame, evidence, the-doctrine, buyers-checklist, lane-discipline, watchlist, about, or mcp.

Searches the published framework and returns matching sections grouped by page.

Arguments: query; optional limit from 1 to 20.

get_full_framework

Returns the complete llms-full.txt document.

Arguments: none.

Connect a compatible client

Use the remote server URL:

{
  "mcpServers": {
    "decision-grade": {
      "url": "https://decision-grade.ai/api/mcp"
    }
  }
}

Client configuration names vary. In a client that asks for fields directly, use:

Name: decision-grade
Type: HTTP
URL: https://decision-grade.ai/api/mcp

Call the endpoint directly

curl -X POST https://decision-grade.ai/api/mcp \
  -H "content-type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

curl -X POST https://decision-grade.ai/api/mcp \
  -H "content-type: application/json" \
  -d '{"jsonrpc":"2.0","id":2,"method":"tools/call","params":{"name":"search","arguments":{"query":"epistemic distance","limit":5}}}'

Boundaries

Read-only

The endpoint publishes framework content. It does not edit documents, evaluate them, or take action in another system.

Public

No account or API key is required for the public endpoint. Do not send confidential material in tool arguments.

Stateless

Requests do not create a personalized memory or decision record.

The endpoint makes the framework easier to read. It does not verify the framework's claims or make a connected model's answer Decision Grade.

Verify the published behavior

The server source is public in functions/api/mcp.ts.
Tool responses derive from the published full-text bundle and search index.
The server does not call an external language model.
The page source, llms.txt, and llms-full.txt can be compared directly.

If the endpoint and the published files disagree, open an issue.