PDF MCP server for Claude, Cursor, and agents

Connect okraPDF over MCP so agents can resolve PDFs, read structure, query pages, open cited regions, and hand off review decisions.

Add to Claude

Why PDF over MCP

MCP gives agents a tool boundary. okraPDF uses that boundary to expose a PDF as a source with structure, citations, hosted views, and review gates. The model asks for context; okraPDF returns only the relevant page, table, or region with evidence attached.

Remote connector

Use the hosted MCP endpoint from Claude Connectors or bridge it into older local-only clients with mcp-remote.

{
  "mcpServers": {
    "okrapdf": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://api.okrapdf.com/mcp"]
    }
  }
}

Tool shape

Resolve PDF URLs or uploaded documents into reusable sources.
Read document structure before sending a full page to a model.
Ask questions with cited page and region evidence.
Open the source or review portal in the browser for human handoff.
Reuse the same source from CLI, SDK, and API jobs.

Where it fits

Use PDF MCP for Claude, Cursor, Codex, internal agents, and workflows that need deterministic document tools. For backend automation, use the same verbs through the okra CLI or REST API.

Frequently asked questions

What is a PDF MCP server?

It is a Model Context Protocol server that exposes PDF operations as tools an AI client can call. okraPDF focuses on structured context, citations, hosted views, and review handoffs.

Does okraPDF MCP work with Claude?

Yes. Use the remote connector URL in Claude Connectors, or use mcp-remote for older clients that only support local stdio servers.

Is the MCP endpoint different from the API?

The interface is different, but the document substrate is the same. MCP is for agents and interactive clients; the REST API and CLI are better for backend automation.