Process and extract data from PDF files with intelligent caching to reduce redundant parsing. Built for developers who need reliable, production-grade PDF handling in their AI workflows.
io.github.jztan/pdf-mcp
Local install
STDIO
No auth required
How models use it and what it is built for.
Process and extract data from PDF files with intelligent caching to reduce redundant parsing. Built for developers who need reliable, production-grade PDF handling in their AI workflows.
Local install — runs as a subprocess.
Configuration this server reads at startup.
Directory for storing PDF cache (default: ~/.cache/pdf-mcp)
Cache time-to-live in hours (default: 24)
Where to find authoritative docs and source for Jztan Pdf.
Paste any of these into Agent Studio after connecting Jztan Pdf.
Common questions about connecting and running Jztan Pdf.
What does the PDF caching do?
The server caches parsed PDF data to avoid re-processing the same files. You can configure the cache directory (PDF_MCP_CACHE_DIR) and set how long cached data persists (PDF_MCP_CACHE_TTL, default 24 hours).
Where does the PDF cache get stored?
By default, cached PDFs are stored in ~/.cache/pdf-mcp. You can override this by setting the PDF_MCP_CACHE_DIR environment variable to any directory path.
How do I install and run this MCP server?
Install with `uvx pdf-mcp@1.10.0`. The server runs over stdio transport, so configure it in your MCP client (like Claude Desktop) to spawn the process and communicate via standard input/output.
Can I disable caching or change how long PDFs are cached?
Yes. Adjust PDF_MCP_CACHE_TTL (in hours, default 24) to control cache expiration. Set it to 0 to disable caching, or increase it for longer retention of frequently-used PDFs.
What PDF operations does this server support?
The server is production-ready for PDF processing with intelligent caching. For a full list of supported operations (text extraction, table parsing, metadata reading, etc.), refer to the server documentation or test with your MCP client.
MCP Playground runs 10,000+ hosted MCP servers — GitHub, Linear, Notion, Stripe, Sentry and more — across Claude, GPT, Gemini, DeepSeek and 30+ AI models. Compare model answers side-by-side, save agent presets, share runs. Zero install.
Open Agent Studio