OpenFormatDB MCP server

OpenFormatDB

An open directory of file formats: how to recognise each one, which open tools read, write and convert it, how far its structure is documented or reverse engineered, and what a conversion loses — so that nobody, human or AI agent, redoes work that already exists.

browser (nothing is uploaded) or find a chain of tools, on openformatdb.org (site/).

• Humans: browse the directory and format pages, identify a file, and convert it in the

converting or reverse engineering a file: identifyfile, lookupformat, searchformats, findconversion, convertfile, claimformat, publish_findings.

• Agents: query it through the MCP server (packages/mcp) before opening,

Assimp and resvg compiled to WebAssembly, run where the file is (browser or the agent's machine), each downloaded once on first use, pinned and checksummed.

• Engines (packages/engines): ImageMagick, FFmpeg, Pandoc, 7-Zip, GDAL,

Congress FDD, Kaitai Struct, MIME databases, converter format lists…) are aggregated by importers (packages/importers; PRONOM, Wikidata, the File Formats Wiki, Kaitai Struct, converters (FFmpeg, ImageMagick, Pandoc, LibreOffice) and the engines today), with a link back to each source.

• Existing sources (PRONOM, Wikidata, Archive Team's File Formats Wiki, Library of

Sister project of OpenDeviceDB.

See SPEC.md for the design and POLICY.md for what we accept. The record format is schema/format.schema.json.

npm install
npm test                  # unit tests of every package
npm run validate          # check every record against schema/format.schema.json
npm run import -- all     # refresh data/imported/ snapshots (PRONOM_DIR, WIKIDATA_DIR, FFW_DIR, KAITAI_DIR: local copies)
npm run mcp               # MCP server on stdio
npm run build             # static site and JSON API in site/dist (npm run dev: localhost)

Code: AGPL-3.0-or-later (see LICENSE). Data: CC BY-SA 4.0 (see LICENSE-DATA.md).