pdf-reader-mcp

Why we included this project

RAG pipelines and document-grounded agents often stumble on PDFs because plain-text extraction flattens layout and leaves claims unverifiable. Citra, shipped as the @sylphx/citra MCP server, fixes that by returning a structured document model with rows, columns, cell bounding boxes, OCR for scanned pages, and page-level location metadata an agent can cite back to a human. That matters for teams working with financial reports, research papers, or legal documents, where an answer that comes with page and cell coordinates beats a number the model invented. Setup is one npx command with no Docker or API key, and since OCR and extraction run in a native Rust engine, PDFs stay on the local machine. The server also fails closed if the native binary is missing, so production output won't degrade silently.