Use mdkit from Claude and other agents via MCP

mdkit ships an MCP server: agents convert documents to Markdown with the convert_document tool — same limits and credits as the REST API.

Last updated

mdkit exposes its conversion surface as a Model Context Protocol server at https://api.mdkit.online/mcp (streamable HTTP). An agent gets three tools:

  • convert_document — synchronous conversion of a base64-encoded file, returns the Markdown inline.
  • submit_conversion_job — the async lane (API key required); returns a job id.
  • get_conversion_job — poll one of your jobs; small results are inlined, larger ones come back as a result_url to fetch.

Tool calls use the same key and credit tier as REST; tools/list is open for discovery.

Claude Code

claude mcp add --transport http mdkit https://api.mdkit.online/mcp \
  --header "X-API-Key: YOUR_KEY"

Or in a JSON MCP client config:

{
  "mcpServers": {
    "mdkit": {
      "type": "http",
      "url": "https://api.mdkit.online/mcp",
      "headers": {"X-API-Key": "YOUR_KEY"}
    }
  }
}

The key is optional for convert_document (anonymous calls run at the low IP rate limit) and required for the async job tools.

Try it

Try it free.

FAQ

Can Claude read a PDF through mdkit?
Yes. Add https://api.mdkit.online/mcp over the HTTP transport and the agent gets convert_document, which converts a base64-encoded file and returns the Markdown inline.
Does the MCP server need an API key?
Not for convert_document — anonymous calls run at the per-IP rate limit. The async job tools, submit_conversion_job and get_conversion_job, do require a key.
How do I add mdkit to Claude Code?
Run: claude mcp add --transport http mdkit https://api.mdkit.online/mcp --header "X-API-Key: YOUR_KEY". The key is optional for synchronous conversion.