Conversion pipeline

When you paste a docs URL, doc2mcp runs a multi-stage pipeline server-side. No work happens on your machine.

1. Discovery

doc2mcp probes the docs origin for /llms.txt and /llms-full.txt (Mintlify, OpenAI, LangChain, Anthropic, Stripe all expose these). If found, every URL in the manifest is queued.

2. Fetch

For each page, doc2mcp tries ${url}.md and ${url}.mdx first (most modern doc platforms serve raw markdown). HTML scraping is the fallback — and even then, code blocks (<pre><code>), headings, and links are preserved as markdown.

3. Analyze

ASI1 reads the structured docs and extracts:

• API endpoints and methods
• Auth patterns
• Workflows and use cases
• A compressed llms.txt index

4. Build

The MCP server is just a hosted HTTP endpoint at /api/mcp/{projectId}/mcp speaking JSON-RPC 2.0. A unique Bearer token is minted, hashed, and stored.

5. Ready

You get a URL + token. Paste into Cursor. Total time: ~30 seconds to a few minutes depending on docs size.

Supported source formats

Paste any of these and doc2mcp picks the right strategy automatically:

Auto-detection

doc2mcp picks the right handler based on:

1. URL extension (.md, .mdx, .json, .yaml, .yml)
2. Hostname (github.com, raw.githubusercontent.com)
3. Path hints (/openapi, /swagger, /spec)
4. Content sniffing once fetched (JSON keys like "openapi": or "swagger":)
5. Marketing-page filter — if you paste stripe.com, doc2mcp redirects to docs.stripe.com automatically