maxjendrall/read_pdf_tool

read_pdf_tool is a command-line program that converts a PDF file into clean Markdown text and one PNG image per page. The main goal is to give an AI agent both the extracted text and visual snapshots so it can answer questions about the PDF accurately. By default it uses a service called Mistral OCR, and Azure Document Intelligence is available as an alternative provider. To get started you need a tool called uv and a Mistral API key. You install the program with a single command, then run an interactive setup step that asks for your API key. The key is stored locally in a private file. Once configured, you run the program against a PDF and it prints the location of the generated files. You can also run it without installing by using uvx, which fetches the package on the fly. For each PDF the tool creates a timestamped output directory containing several files. There is a combined Markdown file with all pages, individual Markdown files per page, PNG snapshots for visual checking, a raw provider response, and a metadata file. Common options let you write the Markdown to a specific path, process only a range of pages, skip image generation, or print machine-readable results for other programs to consume. The intended workflow for agents is to run the tool, read the Markdown path from the summary, and inspect every PNG snapshot unless visual verification is unnecessary. If the tool fails, a text-only fallback called markitdown can be used, though the README notes it may produce lower-quality output. Configuration supports environment variables that override the local config file. The README also covers cost and privacy: PDF contents are sent to the chosen OCR provider, so you should not process documents you cannot share. Mistral OCR is paid, so checking current pricing is recommended. Troubleshooting guidance and local development instructions are included as well.