pdf-reader-mcp

PDF Reader MCP is an MCP server that gives AI agents PDF processing through a single `read_pdf` tool. You pass a list of sources, each a local path or URL with optional page ranges such as `"1-5,10,15-20"`, and flags to include full text, metadata and page count. The README says it extracts text, images and metadata, processes pages in parallel for a claimed 5-10x speed-up over sequential reading, orders content by Y-coordinate to preserve layout, and isolates errors per page so one bad page does not fail the whole file.

The README reports 94%+ test coverage and 103 passing tests, and quotes throughput figures such as 5,575 ops/sec for text extraction; these are the project's own benchmarks. Install is `npx @sylphx/pdf-reader-mcp`. The Security and Sandboxing section is worth reading: by default the server can read any local file the host process can access and fetch any HTTP(S) URL, and you should restrict it with `--allow-dir` flags or the `MCP_PDF_ALLOWED_DIRS` variable, which checks relative paths, absolute paths and `..` traversal after resolution.

Note that the same authors' newer project, anymd, lists itself as formerly pdf-reader-mcp and adds more formats, so check the anymd review if you need more than PDFs. The licence is MIT and the last push was 2026-10-11. This review is based on the README and repository data; we have not run it ourselves.