PageIndex MCP: long PDFs for an agent
PageIndex MCP
An MCP server that lets an agent read long PDFs through a document tree: retrieval by reasoning instead of a vector database
- 406 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- The local tool reads a file at a path the agent names and sends it to the PageIndex cloud service
- The API key sits in the client configuration as a request header
Install
- Works in
Manual install
npx -y @pageindex/mcpA local server with local PDF upload; Node.js 18 or newer is required.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The local tool reads a file at a path the agent names and sends it to the PageIndex cloud service
- The API key sits in the client configuration as a request header
This is third-party code. Review the repository files before installing.
What it does
PageIndex MCP connects an agent to PageIndex, a system that turns a document into a hierarchical tree of sections and finds what is needed by multi-step reasoning. Analyzing a long PDF needs no vector database, no chunking and no preset number of top matches. The agent walks the document structure the way a person uses a book's table of contents and shows a search path that can be checked. There are three ways to connect. A developer takes an API key from the PageIndex dashboard and sets the remote address https://api.pageindex.ai/mcp with an Authorization header. A PageIndex app user signs in over OAuth through the app address or installs the ready .mcpb package in Claude Desktop with a double click. For local files you run your own server with `npx -y @pageindex/mcp`. The local server adds a process_document tool: it takes a PDF link or a file path, uploads the document to the service and submits it for processing. The server takes the remaining tools from the remote PageIndex side.
Who it is for. For people who work with long PDFs in Claude, Cursor or their own SDK-based agent and hit the context limit when reading a whole document.
Good fit when
- A long report, contract or book does not fit in the context window and the agent reports hitting the limit
- You want answers with a transparent, checkable search path through the sections of a document
- You build an agent on the Claude Agent SDK, Vercel AI SDK, OpenAI Agents SDK or LangChain and need pluggable document retrieval without a vector database
- The files sit on your computer and the server must accept a local PDF or a document by link
- You want one way of working with PDFs across several clients, from Claude Desktop to your own code
Not a fit when
- The document must not leave your machine: the file is uploaded to PageIndex for processing
- You need formats other than PDF: the upload tool accepts only PDF
- You need to work offline: the server talks to the remote PageIndex side
- You want a retrieval stack on a vector database fully under your control: PageIndex is built as a hosted service
Example request
Upload ~/Documents/annual-report.pdf to PageIndex and find the section on liquidity, with page numbers.MCP
- Transport
- stdio, http
- Authentication
- OAuth
Limitations
A PageIndex account is required: either an API key from the developer dashboard or a sign-in to the app over OAuth. 1000 pages are free, and the service terms apply beyond that. Running the local server needs Node.js 18 or newer.
The upload tool processes PDF only. The document goes to the PageIndex server, so check that sending a confidential file is acceptable before you do. Retention periods and data handling are described on the service site and are not in the repository. Access to the service from Russia was not checked for this card.
How to disable. Remove the pageindex entry from your MCP client configuration. Uninstall the .mcpb package on the extensions page of Claude Desktop. Revoke the API key in the PageIndex developer dashboard.
FAQ
How do I connect PageIndex MCP to Claude Desktop?
Download the .mcpb file from the repository's releases page and open it with a double click; the OAuth sign-in happens by itself. For other clients add the address https://app.pageindex.ai/mcp to the configuration.
How is PageIndex different from ordinary RAG on a vector database?
The document becomes a tree of sections and the model searches it by reasoning, like a book's table of contents. There is no chunking and no vector database, the number of top matches is not preset, and the search path can be traced.
Do I need an API key?
Connecting from code and SDKs needs a key from the PageIndex developer dashboard, passed in the Authorization header. If you have a PageIndex app account, the client signs in over OAuth without a key.
Can I read a local PDF?
Yes, run the local server with npx -y @pageindex/mcp. The process_document tool takes a file path or a link, uploads the document to the service through a signed address and submits it for processing. The destination folder can be set with the folder_id parameter.
Which frameworks does PageIndex MCP work with?
Besides Claude and Cursor, connection is described for the Claude Agent SDK, Vercel AI SDK, OpenAI Agents SDK and LangChain. Any client with MCP support will do.
Discussion
No questions or reviews yet.
Related
MCP serversMCP PDF for large documents: the agent searches by meaning and keyword across one file or a folder and reads only the pages, tables and scans it needs
An MCP server built on Docling: converts PDFs into structured documents, builds new ones and feeds them to RAG
A PDF-reading MCP server: structured text, tables, OCR, visual crops and page-level citations
MCP PDF from ComPDF: the agent converts PDFs and images to Word, Excel, Markdown and JSON and merges, splits, protects and compares files via a cloud API
An MCP server that turns PDF, DOCX, XLSX, PPTX, images, audio, web pages and YouTube transcripts into Markdown
A local document base for your agent: hybrid semantic and keyword search plus a web UI for browsing and uploads
Comparison with similar tools
Entries that share the main search query, and how they differ.
| Entry | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| PageIndex MCP: long PDFs for an agent (this entry) | yes | OAuth | 406 | |
| pdf-mcp | no | not required | 162 | local search by meaning and keyword inside one file or folder |
| Citra | yes | not required | 1,058 | PDF reading with tables, OCR and page-level citations |
| MCP Local RAG | no | not required | 425 | local document search on a vector database with no keys |
README in short
The server is written in TypeScript. The repository has a manifest.json for Claude Desktop and a server.json for the MCP registry, and connection is described in three options with ready configuration blocks.