AI assistants are great at reading PDFs — and surprisingly helpless at doing things to them. Ask Claude to merge three PDFs or stamp DRAFT across a document and, out of the box, it can only apologize. The fix is the Model Context Protocol (MCP): an open standard that lets assistants call external tools. pdfia runs a free public MCP server, so any MCP-capable agent gets a full PDF toolkit with one line of configuration.
What your agent gets
Connecting the pdfia server gives your assistant ten PDF operations it can call directly: merge, split, extract pages, remove pages, rotate, watermark, add page numbers, read metadata, write metadata, and flatten forms. Files are passed in the tool calls themselves and processed in memory — nothing is stored, and there's no API key or account.
Connect it in 30 seconds
Claude Code
One command:
claude mcp add --transport http pdfia https://pdfia.com/mcpClaude Desktop, Cursor, and other MCP clients
Add the server to your client's MCP configuration:
{ "type": "http", "url": "https://pdfia.com/mcp" }The server is stateless HTTP ("Streamable HTTP" in MCP terms), so there's nothing to install, no Node or Python process to babysit, and it works from any machine.
Then just ask
Once connected, PDF work becomes conversation:
- "Merge these three PDFs into one file, in this order."
- "Watermark every page of contract.pdf with CONFIDENTIAL."
- "Pull pages 4–9 out of this report into a new PDF."
- "Strip the author metadata from this document before I send it."
- "Rotate the second page 90 degrees and add page numbers."
The agent picks the right tool, sends the file, and hands you back the result — no tab switching, no manual uploads.
Limits and privacy
The free tier allows files up to 10 MB, 20 files per request, and roughly 30 operations per 10 minutes per IP — generous for personal and agent use, tight enough to keep the service free for everyone. Processing happens in memory at the edge and files are never written to disk. For maximum-sensitivity documents, the browser tools remain the gold standard: those never leave your device at all.
For developers: there's a plain REST API too
Everything the MCP server does is also available as JSON endpoints —POST https://pdfia.com/api/v1/merge, /watermark, and so on, with files as base64. The developer docs cover every operation, andllms.txt describes the whole site in an agent-readable format. The server is also listed in the official MCP registry as com.pdfia/pdf-tools.
Frequently asked questions
Does it work with agents other than Claude?
Yes — MCP is an open standard. Cursor, Windsurf, VS Code's agent mode, and a growing list of frameworks (LangChain, OpenAI Agents SDK adapters, and others) can all consume MCP servers.
Why is it free?
The operations are lightweight and run at the CDN edge, so costs stay small. If heavy usage ever requires a paid tier, the free tier described here stays.