@writechoice/fumadocs-mcp

MCP (Model Context Protocol) server route handler for Fumadocs projects. Exposes your documentation as an MCP endpoint that AI assistants can query directly — no separate indexing step, no static JSON file. It reads from the Fumadocs source API you already have.

Installation

npm install @writechoice/fumadocs-mcp

Setup

1. Add to next.config.ts

const config: NextConfig = {
  transpilePackages: ["@writechoice/fumadocs-mcp"],
  // ...rest of your config
};

2. Create the MCP route

// src/app/mcp/route.ts
import { createMCPHandler } from "@writechoice/fumadocs-mcp";
import { source } from "@/lib/source";

export const { GET, POST, OPTIONS } = createMCPHandler(source, {
  server: { name: "My Docs", version: "1.0.0" },
});

That's it. Your MCP server is live at /mcp.

Options

createMCPHandler(source, {
  server: {
    name: string;    // displayed in GET /mcp and MCP initialize response
    version: string; // semver string
  };
});

Endpoints

| Method | Path | Description | | --------- | ------ | ---------------------------------------- | | GET | /mcp | Server info and capabilities (tool list) | | POST | /mcp | MCP JSON-RPC 2.0 handler | | OPTIONS | /mcp | CORS preflight |

MCP Tools

list_docs

Lists all documentation pages grouped by section, with title, URL, and description.

{ "method": "tools/call", "params": { "name": "list_docs", "arguments": {} } }

get_doc

Fetches the full processed markdown content of one or more pages by URL path. Accepts a single path or an array for batch reads.

{
  "method": "tools/call",
  "params": {
    "name": "get_doc",
    "arguments": { "path": "/api-reference/charges/create-charge" }
  }
}

{
  "method": "tools/call",
  "params": {
    "name": "get_doc",
    "arguments": { "path": ["/documentation/introduction", "/api-reference/get-started/authentication"] }
  }
}

search_docs

Fast keyword search over page titles and descriptions. Good for finding a page by name.

{
  "method": "tools/call",
  "params": {
    "name": "search_docs",
    "arguments": { "query": "webhook", "limit": 5, "scoreThreshold": 0.1 }
  }
}

search_docs_fulltext

Full-text search using the Fumadocs Orama index. Searches page bodies, headings, and paragraphs. Use this when search_docs returns no results or you need to find mentions within content.

{
  "method": "tools/call",
  "params": {
    "name": "search_docs_fulltext",
    "arguments": { "query": "idempotency key", "limit": 10 }
  }
}

Testing

With your dev server running:

# Server info
curl http://localhost:3000/mcp

# Initialize MCP session
curl -X POST http://localhost:3000/mcp \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{}}'

# List all docs
curl -X POST http://localhost:3000/mcp \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":2,"method":"tools/call","params":{"name":"list_docs","arguments":{}}}'

# Full-text search
curl -X POST http://localhost:3000/mcp \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":3,"method":"tools/call","params":{"name":"search_docs_fulltext","arguments":{"query":"charges"}}}'

How it works

createMCPHandler closes over your Fumadocs source instance and uses it directly:

| MCP Tool | Fumadocs API | | ---------------------- | ---------------------------------------------------------------- | | list_docs | source.getPages() | | get_doc | source.getPage(slugs) + page.data.getText('processed') | | search_docs | Keyword scoring over page.data.title / page.data.description | | search_docs_fulltext | createFromSource(source) — same Orama index as /api/search |

No filesystem walking, no static JSON index. Content is always in sync with your docs at build time.

Requirements

• fumadocs-core >= 16
• next >= 15
• source.config.ts must have postprocess: { includeProcessedMarkdown: true } for get_doc to return full content