PDF to Markdown — MCP setup

Name: PDF to Markdown
Author: dudko

Connect your AI assistant to pdf2md — PDF to clean Markdown, tables and images included — through our hosted Model Context Protocol (MCP) server. Sign in once with your dudko.dev account; no API key to paste.

The server is the front door to the pdf2md HTTP API, not a converter: a document never travels through the model or through the server. The assistant asks for a request, and runs it wherever the PDF is.

Server: pdf-to-md
Transport: streamable HTTP
URL: https://pdf2md.dudko.dev/mcp
Auth: OAuth 2.1 via auth.dudko.dev — you'll be prompted to sign in.

What you can do

Two tools, both free:

Tool	What it does
`api_access`	How the pdf2md API works: endpoints, options, limits, errors.
`api_request`	Checks conversion options (`convert`, `detect`, `text`) and returns the exact request — URL, headers, a curl command — with a 15-minute credential for your account.

Install in Claude Code (plugin)

/plugin marketplace add dudko-dev/pdf-to-md-mcp-setup
/plugin install pdf-to-md@pdf-to-md
/reload-plugins

Then run /mcp and complete the sign-in when prompted.

Prefer not to use the marketplace? Add the server directly:
claude mcp add --transport http pdf-to-md https://pdf2md.dudko.dev/mcp

Connect from other clients

Same remote URL, same sign-in — only the config location differs.

Claude Desktop / claude.ai connectors

Open Settings → Connectors → Add custom connector.
Name: PDF to Markdown
URL: https://pdf2md.dudko.dev/mcp
Save, then connect and sign in when prompted.

Cursor

~/.cursor/mcp.json (global) or .cursor/mcp.json (per project):

json

{
  "mcpServers": {
    "pdf-to-md": {
      "url": "https://pdf2md.dudko.dev/mcp"
    }
  }
}

VS Code (GitHub Copilot / MCP) and other clients

json

{
  "servers": {
    "pdf-to-md": {
      "type": "http",
      "url": "https://pdf2md.dudko.dev/mcp"
    }
  }
}

Authentication

The server is an OAuth 2.1 resource server. The first call answers 401 with a WWW-Authenticate header pointing at its protected-resource metadata:

https://pdf2md.dudko.dev/.well-known/oauth-protected-resource/mcp

which names https://auth.dudko.dev as the authorization server (scope usage). Compliant MCP clients discover this automatically — you only need the server URL, and you sign in with your dudko.dev account. Tokens are audience-bound to https://pdf2md.dudko.dev/mcp and are refused anywhere else.

The server itself carries no files and no content: its tools return a ready-to-run HTTP request with a short-lived credential (15 minutes, good for this API only) for the account you signed in as. Your assistant runs that request wherever the file is, and the API answers the caller directly. Each request counts against your account's plan like any other API call; the MCP tools themselves are free.

Troubleshooting

401 / asked to sign in again — the token expired; reconnect in the client.
"plan used up" / 402 — the conversion quota for your account is spent; retrying will not help. See https://auth.dudko.dev/account.
422 from /convert — the PDF is a scan with no text layer. pdf2md does no OCR; detect tells you this before you spend a conversion.
Tools don't appear — check the client finished the sign-in (/mcp in Claude Code shows connected).