TL;DR: gemini-faf-mcp now carries a real agent.fafa passport, an MCP Server Card, and an AI Catalog entry — built from the server's actual tools, not hand-typed. GEMINI.md also got genuinely richer. A2A is explicitly not in this release — we don't claim it until it's real.

In Plain English

Old state. gemini-faf-mcp had a name and a version number in the registry, but nothing that proved what it actually was — no passport, no verifiable card.

Fix. agent.fafa is now built by reading the server's real code — its actual 12 tools, one by one — not guessed or copied from another server's tool list. An MCP Server Card and an AI Catalog entry ship alongside it.

New state. The registry entry, the catalog entry, and the card all say the same true thing about the server, because they're all built from the same source.

uvx gemini-faf-mcp

uvx is the try. Zero-install, fetched straight from PyPI. For a permanent install: pip3 install gemini-faf-mcp@2.6.0.

What happened since The Dart Edition

One blog, one arc. Last write-up was v2.5.0 — The Dart Edition (Dart/Flutter detection). Short tour to today:

  • 2.5.1 — one.faf namespace migration: joined the fleet on one.faf/gemini-faf-mcp. Identity housekeeping, not a blog.
  • 2.6.0 — The Agent Card Edition. This is the log entry for the stretch.

What 2.6.0 is

Lesson: a registry listing is a claim. A card is proof.

agent.fafa is authored from live introspection of the server's own code — every tool's real name and real docstring, read directly from server.py. It ships next to an MCP Server Card (SEP-2127, carries the FAF context block by construction) and a fresh AI Catalog entry (.well-known/ai-catalog.json — the spec was only ratified in November 2025).

GEMINI.md also stopped being a stub. It now renders real setup and test commands and a "where things live" section — the same richness AGENTS.md already had, brought over to match.

What's not in this release: A2A support. The tooling that builds these cards correctly refuses to fabricate an A2A endpoint this server doesn't serve. Real A2A support is planned as its own future release — not bundled in early to make this one sound bigger than it is.

Try it

Try (zero-install, not the permanent install):

uvx gemini-faf-mcp

Install (PyPI):

pip3 install gemini-faf-mcp@2.6.0

Add to Gemini CLI:

gemini extensions install https://github.com/Wolfe-Jam/gemini-faf-mcp

More depth: Dart/Flutter detection (v2.5.0) · repo · PyPI.

Technical details

  • Version: 2.6.0 (August 20, 2026)
  • Identity: one.faf/gemini-faf-mcp
  • Framework: FastMCP 3.4.7
  • Tools: 12 — unchanged this release
  • Tests: 243 passing
  • Score: Trophy 100%, 17/17 slots
  • New: agent.fafa · MCP Server Card (SEP-2127) · AI Catalog entry
  • Not yet: A2A support — a real future release, not this one