TL;DR: claude-faf-mcp now speaks every AI format. Define your project once in project.faf, and the MCP server generates CLAUDE.md, AGENTS.md, .cursorrules, and GEMINI.md — all without touching a CLI. 5 new tools, 7 bundled parsers, 351 tests passing.
What's New
AI Format Interop
One project.faf. Every AI instruction format. Generated inside Claude Desktop or Claude Code — no terminal required.
5 New MCP Tools
- faf_agents — Import/export/sync AGENTS.md (OpenAI Codex, Linux Foundation standard)
- faf_cursor — Import/export/sync .cursorrules (Cursor IDE)
- faf_gemini — Import/export/sync GEMINI.md (Google Gemini CLI)
- faf_conductor — Import/export Google Conductor directory
- faf_git — Generate .faf from any public GitHub repo URL
Bi-Sync All At Once
One call generates every format simultaneously:
faf_bi_sync { all: true }CLAUDE.md + AGENTS.md + .cursorrules + GEMINI.md — synced in milliseconds.
GitHub Context Extraction
Point at any public repo, get instant AI context:
faf_git { url: "https://github.com/owner/repo" }No clone. No install. 2 seconds to 90%+ AI-readiness.
100% Standalone
Zero CLI Dependency
7 bundled parsers. All format logic runs inside the MCP server. No faf-cli required.
The parsers — agents-parser, cursorrules-parser, gemini-parser, conductor-parser, github-extractor, faf-git-generator, slot-counter — are all bundled directly. Same architecture that made v3.0.4 standalone, now extended to every AI format.
README Upgrade
The npm page is the storefront. v4.5.0 ships with a restructured README:
- 6 Ws Quick Reference — WHO/WHAT/WHERE/WHY/WHEN/HOW at the top
- Navigation bar — Jump to any section
- DAAFT economics — 91% token waste, $5,460/year per developer
- CLI vs MCP table — Same format, different execution layer
Try It
npm install -g claude-faf-mcp Or add to your MCP config:
The Numbers
- v4.5.0 — Released February 27, 2026
- 351/351 — Tests passing (95 new)
- 61 — MCP tools (25 core + 36 advanced)
- 100% — Trophy score
- 27,000+ — npm downloads (ecosystem)