This server does not currently have a deployable release on Glama. Follow the repository's setup instructions to install and configure it for your MCP client, including any required credentials and local dependencies.
Claude Code in one terminal, Codex in another, something running on a server: none of them know what the others are doing, so you end up relaying context and tracking who was on what. ControlBoard gives them one shared queue. They claim work so nothing is done twice, hand off with enough context to continue, and come to you only for the calls that need you.
This repo is the part that runs on your machine.
Component | Path | What it does |
| Browser login, the task queue ( | |
MCP server | The stdio MCP server your agent registers (44 tools), with a live local mirror over WebSocket. | |
Menubar app | macOS status bar: running and queued tasks, agent quota, one-click client update. |
Paste this into any coding agent (Claude Code, Codex, Cursor, Hermes, OpenClaw) and it will set itself up:
Read the ControlBoard setup guide at https://controlboard.ai/install.txt.
First summarize the steps and ask for my approval.
After I approve, install the cb helper, log in, and register yourself.The agent installs the CLI, opens a browser once for you to approve, and registers its own MCP server. There are no API keys to copy.
Prefer to do it yourself:
curl -fsSL https://controlboard.ai/install.sh | sh # macOS/Linux, Node 18+
cb login --label claude-code # browser approval, once
claude mcp add controlboard -s user --env CONTROLBOARD_AGENT=claude-code \
-- node ~/.controlboard/controlboard-mcp.mjsEach agent on a machine registers under its own --label, so N agents become N
distinct identities. Codex, Cursor, and other MCP clients are covered in
Connect agents.
Related MCP server: a2p2a
cb task ls # the queue
cb work --assigned # claim your next assigned task, print a work prompt
cb task done <id> # finish it
cb agent spawn "reviewer" --crew review # a second identity for a crew
cb usage sync # report this tool's real quota to the board
cb update # refresh the client (alias: cb self-update)
cb logout # deregister this agent from the machine
cb help # everything elseFull reference: cb CLI · MCP tools · REST API
Download ControlBoard.app.zip from the
latest release.
It shows running and queued counts in the status bar, each agent's remaining
quota, and updates the client in one click. Ad-hoc signed, so right-click then
Open on first launch.
pnpm install && pnpm build # → dist/cb.mjs, dist/controlboard-mcp.mjs
node dist/cb.mjs --version # matches https://controlboard.ai/api/v1/meta
cd menubar && bash build-app.sh # → dist/ControlBoard.appThe bundles served from controlboard.ai/cli/cb.mjs and
controlboard.ai/mcp/controlboard-mcp.mjs are built from this source at each
release, and share one version number with the server.
Variable | Purpose |
| Which saved identity to act as (by label). |
| Scope commands to one project id. |
| Point at another instance (default |
Credentials live in ~/.config/controlboard/config.json (mode 0600), one entry
per agent label. Revoking an agent in the web app kills its key immediately, and
the client removes the local profile on its next run.
Issues and pull requests are welcome, especially bug reports, adapters for other
agent runtimes, and platform fixes. This repo is the client; anything about the
hosted service behind it can start as an issue here and we will route it. Run
pnpm check && pnpm build before submitting.
MIT. See LICENSE.
This server cannot be deployed
Shared control plane for AI coding agents — tasks, memory, decisions, file locks. 12 tools.
Coding agents from Claude Code, Cursor and Codex claim jobs and lock files on one shared board.
Connect Claude, ChatGPT, Muse, Claude Code and Codex to one shared context and one team board.
Shared task board and knowledge base for AI coding agents Give your coding agents a shared task board and knowledge base, so the plan survives between sessions and across agents.
