MCP server

Let AI agents read scanned documents as a tool.

docbeast has a remote Model Context Protocol server, so AI clients can convert documents. The server URL is:

https://mcp.docbeast.com/mcp

Connect

Without signing in, a client can use the free conversions: the first page of a file, 3 times a day. There's nothing to set up.

With OAuth, add the URL as a custom connector in your client. It sends you to docbeast to sign in with your email link, and you choose Allow. You never paste a key. To disconnect a client later, use Connected apps on the dashboard.

With an API key, for clients that let you set headers:

{
  "mcpServers": {
    "docbeast": {
      "url": "https://mcp.docbeast.com/mcp",
      "headers": { "Authorization": "Bearer dbk_live_..." }
    }
  }
}

Account keys and credit keys both work. See Authentication.

Tools

Tool	What it does	Counts pages
`convert_document`	Reads a PDF or image and returns its tables and download links.	Yes
`get_conversion`	Reads a conversion's status and, when finished, its tables and download links.	No
`get_usage`	Shows the pages used and remaining. Needs a sign-in or key.	No

convert_document

Argument	Meaning
`url`	An `https` link to the document. Give this or `base64`, not both.
`base64`	The document's bytes, base64-encoded.
`filename`	A name for the file, used to name the downloads.
`mode`	`table` (default) keeps rows and columns. `text` reads the page as plain text.

The tool waits up to 45 seconds. If the conversion is still queued or running after that, it returns the job's id and status, and the client can follow up with get_conversion. The conversion keeps going either way.

The result is the same job as the REST API returns, with the tables and text added as pages. Only the first 200 rows are included, and the download links have everything.

Limits

MCP uses the same plans and limits as the API, and pages are counted the same way. Files can be up to 25 MB.

Some problems are returned as a tool result with isError and a code:

Code	Meaning
`bad_input`	Give exactly one of `url` or `base64`.
`bad_url`	The link isn't valid or isn't `https`.
`fetch_failed`	Downloading the document failed.
`bad_base64`	The `base64` value couldn't be decoded.
`file_too_large`	The document is over 25 MB.
`unsupported_file`	The file isn't a supported type.
`limit_reached`	The account or key has no pages left.
`not_found`	There's no such conversion, or it has expired.

Sign-in problems are HTTP errors instead, so the client can start OAuth. A credential that's invalid, expired or revoked gets a 401 on any message, and a call without a credential gets a 401 once today's free conversions are used.

Discovery

URL	What it is
`https://mcp.docbeast.com/.well-known/mcp.json`	Server name, tools and links, for registries.
`https://mcp.docbeast.com/.well-known/oauth-protected-resource`	OAuth protected resource metadata.
`https://docbeast.com/.well-known/oauth-authorization-server`	OAuth authorization server metadata.