pdfmux MCP
pdfmux is a self-healing PDF extraction tool that audits its own output and certifies other extractors. It routes pages to the best of 7 backends or BYOK LLMs, providing per-page confidence scores and structured JSON/Markdown output. The native MCP server exposes 7 tools for conversion, analysis, metadata retrieval, batch processing, structured extraction, streaming, and verification.
Package and access
Local package with source-attributed installation metadata. Koskamo does not execute packages during review.
Checks and signals
Each line is separate evidence, not a trust score.
- Popularity 86 GitHub stars
What you can do with it
- Extract PDFs to Markdown with self-healing quality checks
- Verify other extractors' output for silently dropped pages
- Stream NDJSON page events for long documents
- Extract structured tables and key-values to JSON
Source and maintenance
- Last repository push
- Latest release
- v1.8.7
- Maintenance checked
Packages and installation
Source-provided package data. Koskamo has not executed or live-probed this package.
pdfmux 1.8.7
{
"version": "1.8.7",
"transport": "stdio",
"identifier": "pdfmux",
"runtimeHint": "pdfmux",
"registryType": "pypi",
"runtimeArguments": [
{
"type": "positional",
"value": "serve",
"isRequired": true
}
],
"environmentVariables": [
{
"name": "GEMINI_API_KEY",
"format": "string",
"description": "Optional Google AI API key. Enables the paid Gemini Flash fallback for pages where all local backends report low confidence. pdfmux runs 100% locally without this."
}
]
}Tools
Reported by the attributed source; not independently observed by Koskamo.
Maintenance history
Public check and source-refresh events. Editorial and ownership records remain private.
- · repository · passed · metadata changed · updated
- · backfill · recorded · Pre-expansion baseline; unobserved rich metadata remains unknown
Let your agent find its tools
Find servers like pdfmux MCP from your agent. Add Koskamo once to search the index for endpoints, access requirements, and client setup. Public discovery needs no key.