What it does
This MCP server extracts text, images, and metadata from PDF files, emphasizing parallel processing for performance. It orders content by Y-coordinates to preserve document layout, supports page-range selection (e.g., "1-5,10,15-20"), and handles both absolute and relative paths across Windows and Unix systems. Built in TypeScript with 173 tests and 94%+ code coverage, it claims 5–10x speedup over sequential processing for multi-page documents.
Who it's for
Backend engineers building document analysis features into Claude workflows, data scientists preprocessing PDFs for RAG pipelines, and DevOps teams automating PDF extraction in batch workflows where sequential I/O is a bottleneck.
Common use cases
- Extract full text and metadata from PDF documents for AI-powered content analysis
- Process specific page ranges from large manuals or reports to reduce token usage
- Batch-process 50+ page documents in parallel to reduce end-to-end latency
- Preserve document reading order via Y-coordinate sorting when layout semantics matter
- Validate PDF structure and metadata before passing to downstream analysis systems
Setup pitfalls
- CI tests are currently failing (as of today's commit) — verify the build status and changelog before production deployment
- Filesystem read/write access required — sandbox this server if processing untrusted or user-supplied PDFs
- Windows path support was added in v1.3.0; earlier versions expect Unix-style paths only
- Tool discovery may show zero tools despite functional operations — cross-reference the README or use
mcp inspectto verify capabilities