MCP - Model Context Protocol#
TetraFi docs ship as an MCP server. Any MCP-aware agent - Claude Desktop, Cursor, Windsurf, Continue, or your own - can search, read, and introspect the corpus the same way your IDE autocomplete does.
The manifest lives at /docs/mcp/manifest.json. Point any MCP client at that URL to discover the full tool set.
What you get#
Enumerate every doc page with slug + metadata.
Raw markdown + frontmatter for any slug.
Ranked substring search across the corpus.
Resolve an OpenAPI schema by name.
Resolve an OpenAPI operation by id or method+path.
Install - Claude Desktop / Cursor#
Native MCP clients speak JSON-RPC over stdio. Drop this thin bridge into your claude_desktop_config.json (or Cursor's mcp.json):
1{2 "mcpServers": {3 "tetrafi-docs": {4 "command": "npx",5 "args": ["-y", "mcp-remote", "https://tetrafi.io/docs/mcp/manifest.json"]6 }7 }8}Restart the client. tetrafi-docs appears as an available tool set; the assistant can now call search_docs, read_doc, and the rest.
No MCP client? The same endpoints work as plain REST - curl https://tetrafi.io/api/mcp/search-docs?q=rfq returns JSON.
Direct REST examples#
1# List every published doc2curl https://tetrafi.io/api/mcp/list-docs34# Read a single page as markdown + frontmatter5curl "https://tetrafi.io/api/mcp/read-doc?slug=integration/taker-guide"67# Ranked search8curl "https://tetrafi.io/api/mcp/search-docs?q=webhook%20signature&limit=5"910# Resolve a schema by name11curl "https://tetrafi.io/api/mcp/get-schema?name=Quote"1213# Resolve an endpoint by operationId OR method+path14curl "https://tetrafi.io/api/mcp/get-endpoint?operationId=submitRFQ"15curl "https://tetrafi.io/api/mcp/get-endpoint?method=POST&path=/rfq"