This is a hosted MCP server that converts documentation sites, GitHub repositories, and OpenAPI specifications into callable tools your AI can use. You point it at a docs URL or repo, it ingests the content, and exposes it through the Model Context Protocol over HTTPS. Instead of copy-pasting docs into prompts, you connect via a project-specific token and the AI can query the material directly. It runs remotely at doc2mcp.site, so there's no local setup beyond adding the streamable-http endpoint to your MCP client config. Useful when you're working with large codebases or API references that change frequently and want the assistant to have live access without manual updates.
Paste any docs URL → get a hosted, Cursor-ready MCP server in under 60 seconds.
Live · Docs · SDK · Marketplace · Open Source
⭐ Star this repo — it helps developers discover doc2mcp on GitHub and the MCP Registry.

| Surface | Best for | Start here |
|---|---|---|
| Web | Instant convert in the browser | doc2mcp.site |
| CLI | Terminal + one-shot install into Cursor | npm i -g doc2mcp |
| SDK | Code, CI, LangChain / LangGraph / agents | doc2mcp-sdk |
API keys always come from doc2mcp.site (User PAT + project token). The SDK has no separate login.

Docs URL → crawl → analyze → MCP tools → hosted endpoint → Cursor / Claude / LangChain / LangGraph.
{
"mcpServers": {
"my-docs": {
"url": "https://doc2mcp.site/api/mcp/{project_id}/mcp",
"headers": { "Authorization": "Bearer <project-token>" }
}
}
}
npm install -g doc2mcp
doc2mcp login # creates User PAT (d2mcp_pat_…)
doc2mcp https://docs.example.com
Lists the same projects as the dashboard. Install helpers for Cursor / VS Code / Claude / Windsurf.
doc2mcp-sdk)
npm i doc2mcp-sdk
import { Doc2MCP } from "doc2mcp-sdk";
// User PAT from doc2mcp.site / `npx doc2mcp login` — SDK has no auth of its own
const client = new Doc2MCP({ token: process.env.DOC2MCP_API_TOKEN! }); // Settings → d2mcp_usr_…
const ready = await client.convertAndWait({
sourceUrl: "https://js.langchain.com/docs",
});
const answer = await client.callToolText({
mcpUrl: ready.mcp.url,
mcpToken: ready.mcp.token, // project token d2mcp_…
name: "search_documentation",
arguments: { query: "how do runnables work?" },
});
Full handbook (API keys, Cursor, Claude, LangChain, LangGraph, LangSmith, OpenAI, Vercel AI SDK, CrewAI, AutoGen, Mastra):
Product docs page: doc2mcp.site/docs/sdk
| Tool | What it does |
|---|---|
list_documentation_pages | Every crawled page |
get_documentation_page | Full markdown of one page |
search_documentation | Heading-aware search |
get_documentation_overview | Summary + index |
read_full_documentation | All pages combined |
ask_documentation | Q&A with citations |
| Repo | Purpose |
|---|---|
| doc2mcp/doc2mcp (this) | Next.js app, CLI, pipeline, MCP runtime |
| doc2mcp/doc2mcp-sdk | Programmatic JS/TS SDK + framework examples |
| doc2mcp/doc2mcp-registry | MCP Registry gateway manifest |
staging → open PR → stagingGemini PR review) fails the check on must-fix — fix blockers before mergeopensourcedoc2mcp → Starter free for 12 months on PricingMore: doc2mcp.site/open-source
pnpm install
cp .env.example .env.local
pnpm dev
pnpm check
pnpm exec tsc --noEmit --skipLibCheck
Feature → staging (preview) → main → v* tag deploys production on Vercel.
PR checks: Biome/Ultracite, TypeScript, Next build, CLI build, Gemini AI review (must-fix → failed check).
MIT — see LICENSE.