pdf-reader-mcp

MCP.Pizza Chef: SylphxAI

Now branded Citra, this pulls text, headings, tables, and images out of a PDF sitting on your own computer, and returns the page and position each answer came from so you can check it yourself. Scanned documents go through text recognition instead of coming back as noise. It can search a long document to find the right pages before reading them in full. Nothing is uploaded anywhere and no key is required. It works in Claude Desktop, Claude Code, Cursor, and VS Code.

Files/PDF
Web/Research

Use This MCP server To

Pull a number out of a financial report and see its page Search a long PDF for the pages that mention a topic Read a scanned contract that copy-paste turns into gibberish Get a table out of a report with its rows intact Quote a research paper and know exactly where the quote sits

README

Citra

Give your AI agent eyes for PDFs.

Citra is a local-first PDF evidence product for agents — fast, citeable, owned entirely in this repository.

Turn PDFs into structured text, tables, OCR, visual evidence, and page-level citations — locally — via SDK, CLI, or MCP.

Plain-text PDF tools make agents guess. Citra returns proof.
Package (transition): @sylphx/pdf-reader-mcp · bin pdf-reader-mcp

npm version License: MIT CI stars MCP Toplist

Product docs

Doc Purpose
docs/POSITIONING.md Strategic positioning
docs/COMPETITIVE.md Peer anchors and wedge
docs/EVIDENCE_CONTRACT.md Evidence = result contract
docs/TOOL_SURFACE.md Few clear tools policy
docs/PRODUCT_INDEPENDENCE.md This repo is SSOT
docs/IPPB.md Independent public product bar
docs/PUBLISH.md npm/git publish status

pdf-reader-mcp FAQ

Does my PDF get uploaded somewhere?
No. It reads files on your own machine and does all the work there.
Do I need to pay for a key?
No key and no account. Installing it is the only step.
Can I use this to pull tables out of a report?
Yes. It keeps rows, columns, and cells separate, and tells you which page and which cell each figure came from.
Will it handle a scanned document?
Yes. Scanned pages go through text recognition, and the results still come back tied to a page rather than as one blur of text.
Which apps does it work in?
Claude Desktop, Claude Code, Cursor, VS Code, and other assistants that accept a short settings block.
How hard is setup?
You paste a two-line block into your assistant's settings, or run a single command in Claude Code.