Connect Claude, Cursor, Windsurf and n8n to PDF4's PDF MCP server and let your AI agent convert, OCR, extract, summarize and translate PDFs as native tools — over the Model Context Protocol, with no custom integration code.
/mcp/sse
Claude, Cursor, Windsurf & n8n
Bearer MCP API key
30+ PDF tools
The Model Context Protocol (MCP) is an open standard for connecting AI clients to external tools and data. An MCP server advertises a set of tools that an AI client can discover and call; a PDF MCP server is one whose tools are PDF operations.
PDF4's MCP server exposes document operations — convert, OCR, extract, summarize, translate, merge, split, compress and more — so any MCP-compatible agent can read and transform PDFs directly. Instead of stuffing raw PDF bytes into a prompt or wiring up bespoke HTTP calls, your agent lists the available tools and invokes them as part of a task, getting back clean Markdown, JSON or text it can reason over.
MCP is the same protocol Claude, Cursor and Windsurf use to talk to tools. No proprietary SDK to learn.
The agent calls tools/list to discover every PDF operation and its input schema, then calls tools/call to run one.
Tools return Markdown, JSON and text built for RAG, tool-use and document automation — not broken PDF bytes.
Looking for the broader picture of building agentic PDF workflows — RAG ingestion, function-calling patterns and the REST API? See PDF tools & API for AI agents. This page focuses specifically on the MCP server: connecting it, the tools it offers and how it authenticates.
The PDF4 MCP server speaks the SSE transport at https://pdf4.net/mcp/sse and authenticates with an MCP API key as a Bearer token. Add a server entry to your client's MCP configuration, then restart the client so it discovers the tools.
pdfa_mcp_). MCP server access is included on every paid plan — the REST API is available on the free plan.Open Settings → Developer → Edit Config to edit claude_desktop_config.json, then add the PDF4 server. Many clients accept a remote SSE URL directly:
If your client version does not yet support a remote url, bridge the SSE endpoint with the mcp-remote helper (this shape works for Claude, Cursor and Windsurf alike):
Cursor reads ~/.cursor/mcp.json (or Settings → MCP); Windsurf reads ~/.codeium/windsurf/mcp_config.json. Both use the same mcpServers structure shown above — paste the PDF4 entry, save and reload.
In n8n, add an MCP Client node, point it at https://pdf4.net/mcp/sse and set the Authorization: Bearer pdfa_mcp_YOUR_KEY header. Any MCP-aware automation platform (and agent frameworks that support remote MCP) can connect the same way. For Zapier-style flows without MCP support, call the matching REST endpoints directly.
Note: exact config-file locations and whether a client accepts a remote url vs. requiring the mcp-remote bridge vary by client version — treat the snippets above as illustrative starting points and check your client's MCP docs.
Once connected, the agent calls tools/list to discover every PDF operation and tools/call to run one. The MCP server exposes 30+ tools backed by the same engine as the REST API. The ones agents reach for most:
Plus utility tools like merge_pdfs, split_pdf, compress_pdf, protect_pdf and get_credit_balance. Browse the API docs for the full tool list, input schemas and the MCP section under /mcp/sse.
Tool names are real; the exact argument keys for each tool are defined by its input schema (available via tools/list) — the arguments above are illustrative.
The PDF MCP server authenticates with an MCP API key passed as a Bearer token in the Authorization header (keys are prefixed pdfa_mcp_). This is separate from the JWT used for interactive web sessions, and it is what your client config above carries.
get_credit_balance.Generate a key from your account, drop it into your client config, and the agent is authenticated for every tools/call. Sign up and pick a paid plan to get started — MCP access is included on every paid plan, and the REST API is available on the free plan.
The PDF MCP server and the PDF REST API expose the same tools and share the same key model — pick the surface that fits how you build.
An AI agent or IDE (Claude, Cursor, Windsurf, n8n) should discover and call PDF tools itself. Lowest friction for agentic and tool-use workflows.
You are writing application code, batch jobs or backend services that call PDF operations directly with HTTP and a Bearer token.
Building a full PDF-powered agent — RAG ingestion, function-calling tool definitions, async polling? The AI agents guide walks through the patterns, and the API docs cover every endpoint.
An MCP server is a service that implements the Model Context Protocol, an open standard for connecting AI clients to external tools and data. The server advertises a list of tools an AI client can discover and call. Instead of writing custom integration code, you point an MCP-compatible client like Claude, Cursor or Windsurf at the server and the agent can invoke its tools during a task.
The PDF4 MCP server is a Model Context Protocol server for PDF operations. It exposes tools to convert, OCR, extract, summarize and translate PDFs so any MCP-compatible AI agent can work with documents directly. It runs over SSE at /mcp/sse and is authenticated with an MCP API key passed as a Bearer token.
Add a server entry to your client's MCP config (claude_desktop_config.json for Claude Desktop, ~/.cursor/mcp.json for Cursor, ~/.codeium/windsurf/mcp_config.json for Windsurf) pointing at https://pdf4.net/mcp/sse with your MCP API key as a Bearer token. Clients that accept a remote URL can use it directly; others bridge the SSE endpoint with the mcp-remote helper. Restart the client and the PDF tools appear.
The agent can list tools via tools/list and call them via tools/call. Available operations include convert_pdf, pdf_to_markdown, extract_text, extract_json, ocr_pdf, extract_data, summarize_pdf, translate_pdf, merge_pdfs, split_pdf, compress_pdf and more — each PDF tool also has a matching REST endpoint and tool page.
The MCP server uses an MCP API key sent as a Bearer token in the Authorization header (keys are prefixed pdfa_mcp_). MCP access requires an active paid plan (the REST API works on the free plan); sign up, pick a plan, then generate an MCP API key from your dashboard and put it in your client config. Requests are credit-metered and rate-limited per key.
Use the MCP server when an AI agent or IDE should discover and call PDF tools itself — it is the lowest-friction path for Claude, Cursor, Windsurf and similar clients. Use the REST API when you are writing application code, batch jobs or backend services that call PDF operations directly. Both share the same tools and the same MCP API key model. See the AI agents guide for full agent-building patterns.
Get a free MCP API key, add PDF4 to Claude, Cursor or Windsurf, and give your agent PDF tools in minutes.