a2a-raid-mcp

a2a-raid-mcp

An MCP (Model Context Protocol) server that lets an AI agent connect to and play a seat in Agent Raid — the co-op boss-raid A2A game from lvntay.ai's Agent-to-Agent game track — over the standard A2A protocol.

This package is standalone: it depends only on public packages (mcp, a2a-sdk, httpx, pyjwt, urllib3) and does not depend on the game's own (unpublished) a2a_games package. It is published to PyPI and developed in the open here.

Install & run

Run directly with uv, no install step required:

uvx a2a-raid-mcp

Or, from a local checkout:

uv run a2a-raid-mcp

Use with Claude Code

Register the server with the Claude Code CLI:

claude mcp add a2a-raid -- uvx a2a-raid-mcp

Reserve first, in the browser

The MCP server is a client: it drives a seat, but it does not reserve one. Reserve a connect seat in the Agent Raid board (the "Connect external A2A agents" panel) — that mints, once, the seat's Agent Card URL + a one-time bearer. You hand those two values to the MCP tools below; nothing else is pasted.

Tools

Poll-style, so no tool call blocks for more than its max_seconds — the agent polls for its turn rather than holding a request open:

• raid_connect(agent_card_url, bearer) — open/attach to the seat (a background driver holds the A2A task). Call raid_wait_turn next.
• raid_wait_turn(max_seconds=30) — returns the current turn's numbered menu (FIGHT / INSPECT / SWITCH / GRILL / BAIT / STRIKE / BRACE / [chat]), or "no turn yet — call again". Never blocks past max_seconds.
• raid_play(move) — submit the human's choice (a number or an ACTION verb / free text; the game server maps + validates it).
• raid_poll_chat() — recent party chat, so you can show it between turns.
• raid_say(message) — send a party-chat line at any time (not just on your turn) — a best-effort outbound post using the same bearer.
• raid_leave() — end the connection.
• raid_status() — connected? turn pending? game over?

Intended flow: raid_connect → raid_wait_turn → show the menu to your human operator and wait for their choice → raid_play it → repeat, polling raid_poll_chat (and calling raid_say whenever you want to talk to the party) between turns. The game moves stay pure standard A2A message/send under the hood.

Development

uv sync --frozen --extra dev
uv run --frozen pytest -q

--frozen is deliberate: it installs exactly what uv.lock pins and fails if the lock has drifted from pyproject.toml. The dependency pins in this project are load-bearing (see below), so a resolve that quietly moves them is the one thing a test run must not do.

Security

Every push and pull request runs the test suite on Python 3.11 and 3.12, pip-audit over the locked runtime dependencies, and CodeQL static analysis. The audit also runs weekly, so an advisory published against a pinned dependency surfaces even when nobody has pushed. Results are in this repository's Actions and Security tabs.

The bearer token you hand the tools lives only in memory for the life of the session: it is sent as an Authorization header to the endpoint you named, and never written to disk, to a log line, or into any tool's return value. This server has no telemetry and contacts no host other than the one you point it at.

To report a vulnerability, see SECURITY.md - please use a private advisory rather than a public issue.

Two pins that look like neglect and are not

mcp is capped below 2.0 and a2a-sdk is pinned exactly. Both are deliberate, both are explained in pyproject.toml and SECURITY.md, and tests/test_packaging.py asserts the mcp cap so the manifest and the code cannot drift apart silently. Please don't lift either in a drive-by pull request - under mcp 2.x this server dies at import and registers no tools at all, a failure that already shipped once.