FlightDeck MCP Server

Run more than one AI coding agent on a machine and a familiar problem appears: nobody can say with confidence what each one touched, when it started, whether it is still running, or what it cost. The flightdeck mcp server exists to answer those questions. It dispatches every Claude Code agent onto a single board and records what happens in a hash-chained event ledger, so the history of a session is something you can audit rather than reconstruct from memory.

What the board actually tracks

FlightDeck is built around a small set of governance ideas that show up directly in how it behaves:

• Hash-chained event ledger — each entry is linked to the one before it, so a record cannot be quietly rewritten after the fact.
• Claim adjudication — when two agents both reach for the same work, the board decides who holds it instead of letting them collide.
• HOLD — a way to stop work in place rather than killing a session and losing its context.
• Git-observed scope — what an agent changed is read from git, never taken from the agent's own account of itself. That distinction is the whole point: self-reported scope is exactly the thing you cannot trust when something goes wrong.
• Liveness and cost — whether anything is genuinely running and for how long, with unknown costs recorded as unknown rather than silently counted as zero.

That last detail is a good signal of the project's temperament. Treating an unmeasured cost as zero makes dashboards look tidy and makes them wrong; FlightDeck would rather show you a gap.

Setting it up

As a Claude Code plugin there is a shorter path — add the marketplace, install the plugin, and /flightdeck turns governance on inside a session. Point FLIGHTDECK_BOARD at a folder and you are running; a board is just a local directory, created the first time you make a packet.

Where it fits

Most MCP servers add a capability — a database to query, a browser to drive, an API to call. FlightDeck does something different: it watches the agents you already have. That makes it a poor first server to install and a sensible one to add the moment you are running several agents in parallel and have started losing track of which did what.

It is worth being clear about the trade-off. The package is proprietary rather than open source — free to install, with a patent pending — so it sits differently from the MIT-licensed tooling that surrounds most of this ecosystem. On the other hand, the board is local-first: the board tools send nothing anywhere, and the cloud family stays inert without an operator-issued token, which is a meaningful posture for something observing your engineering work.

Practical notes

• Python 3.9 or newer, with uv on PATH for the recommended install route.
• The ledger is only as useful as your discipline in routing agents through the board; agents started outside it are invisible to it.
• Because scope is read from git, work in a dirty tree or outside a repository will be observed less precisely.
• SyncDev checks each listing against the project's own documentation before publishing, and revisits entries as they change.

More MCP options

Compare MCP servers by category, or browse hosted MCP servers if you prefer an endpoint over a local process.

Maintainer setup documentation.