What it is
pdf-mcp is a command-line tool and MCP server covering most PDF jobs. Its 57 tools fall into groups: form filling and flattening (9), page operations such as merge, split and rotate (8), text tools including regex redaction, watermarks and Bates numbering (13), extraction of text, tables, images and links (7), metadata (4), signing and signature verification (6), OCR via Tesseract, AI tools for auto-filling forms and extracting structured data, and batch processing.
The AI tools use a local vision model by default, then Ollama, and only call OpenAI if you supply a key. It is built on pypdf, fillpdf and PyMuPDF. The repo is Apache 2.0, but PyMuPDF is AGPL-3.0. It has 11 stars; the last tagged release is v1.2.18 from February 2026, though the README documents v1.3 features and commits continue.
Install
git clone, make install, pdf-mcp serveOur review
Invoices, forms and scanned contracts are where office automation usually stalls. pdf-mcp covers the whole range in one install, and the same 57 tools work as CLI commands, so a job an agent figured out can become a script.
Local-first is the right default for documents like these. OCR runs on Tesseract, and the AI features (auto-filling a form from source data, extracting structured fields, classifying a document) try a local vision model, then Ollama, before OpenAI, which needs an explicit key.
Two cautions. Redaction in PDFs is easy to get wrong, because text can survive under a black box or in metadata, so verify any redacted file with a text extractor before sharing it. And PyMuPDF is AGPL-3.0, which matters if you build this into a product you distribute or run as a network service, despite the repo's Apache 2.0 badge.
The project is small, with 11 stars and one maintainer, and the latest tagged release lags the README. The breadth is real; the track record is not yet.
Where it earns its place
A bookkeeper at a 15-person firm with 200 supplier invoices a month
Pull invoice numbers, dates and totals into a spreadsheet
Point the agent at the folder and use extract_structured_data with OCR for scans; the local model keeps documents on the machine and costs nothing per page.
A paralegal preparing a disclosure bundle
Merge exhibits, add Bates numbers and redact personal data
Use merge-pdfs, add-bates-numbering and redact-text-regex, then re-extract the text to confirm nothing redacted survives.
Before you commit
- Setup
- Clone the repo and run make install (uv-based); Tesseract needed for OCR. pdf-mcp serve starts the stdio server.
- Integrations
- Cursor, Claude Desktop and other stdio MCP clients; full CLI for scripts and CI.
- Data and privacy
- Processing runs locally; the OpenAI backend is used only if OPENAI_API_KEY is set. Apache 2.0 code, AGPL-3.0 PyMuPDF.
- Limits
- 11 stars and two open issues; the latest tagged release (v1.2.18) lags the v1.3 features in the README; redaction correctness is your responsibility.
What's good
- 57 tools across forms, OCR, extraction, signing and batch
- Same tools as CLI and MCP server
- AI features default to local models
- Bates numbering and signature verification for legal work
What's not
- PyMuPDF dependency is AGPL-3.0
- Redaction output needs manual verification
- Very small user base; releases lag the README
Best for
People who process invoices, forms or scanned documents and want it done on their own machine.
Alternatives
pdf-mcp: common questions
- Is pdf-mcp worth it?
- The broadest local PDF toolkit we found for agents, with a licence wrinkle and a redaction feature you must check by hand. We rate it 3.5 out of 5.
- How much does pdf-mcp cost?
- Free; Apache 2.0 code with an AGPL-3.0 dependency (PyMuPDF). Optional OpenAI backend bills separately.
- What are the downsides of pdf-mcp?
- PyMuPDF dependency is AGPL-3.0; redaction output needs manual verification; very small user base; releases lag the README.
- Who is pdf-mcp best for?
- People who process invoices, forms or scanned documents and want it done on their own machine.
- Is pdf-mcp safe for company data?
- Processing runs locally; the OpenAI backend is used only if OPENAI_API_KEY is set. Apache 2.0 code, AGPL-3.0 PyMuPDF.
- What are the best pdf-mcp alternatives?
- The closest alternatives we have reviewed are Filesystem (reference server).
Answers checked Oct 24, 2026.