Connect your AI client (MCP)

Briefboard ships an MCP server — a standard Model Context Protocol endpoint that lets tools like Claude Code, Cursor or OpenAI Codex work directly from your project data. No copy-pasting tasks into a prompt: the agent reads and writes the same milestones, tasks, and specs your team does.

Enable Briefboard Dispatch

MCP and the other technical features live behind Briefboard Dispatch. Open your project settings, go to the Dispatch tab and enable it for the project to unlock:

• The MCP server connection
• Specs and version history
• AI Prompt templates
• Custom properties and OAuth clients

Once Dispatch is on, every member has access by default — the member list in the same tab lets you turn it off per member.

The endpoint

The server is exposed as a streamable HTTP endpoint at /mcp/briefboard, protected by OAuth with dynamic client registration. There is nothing to configure server-side: point your client at the URL, approve the authorization screen that opens in your browser, and the connection is saved.

During local development, connect over http:// rather than https://. Some MCP clients use Node's own certificate store and will reject a self-signed certificate otherwise.

Claude Code

claude mcp add --transport http briefboard https://app.briefboard.io/mcp/briefboard

The first time the agent calls the server, Claude Code opens the OAuth authorization screen. Approve it and you are done.

Cursor

Add the server to .cursor/mcp.json in your repository (or ~/.cursor/mcp.json to share it across projects):

{
  "mcpServers": {
    "briefboard": {
      "url": "https://app.briefboard.io/mcp/briefboard"
    }
  }
}

Cursor detects that the server requires authentication: open Settings → MCP and complete the login prompt next to the server.

OpenAI Codex

Declare the server in ~/.codex/config.toml:

[mcp_servers.briefboard]
url = "https://app.briefboard.io/mcp/briefboard"

Then run the OAuth flow once:

codex mcp login briefboard

Other clients

Any client that speaks streamable HTTP can connect with just the URL — OAuth clients are registered dynamically, so there is no manual credential setup. A typical JSON configuration looks like:

{
  "mcpServers": {
    "briefboard": {
      "type": "http",
      "url": "https://app.briefboard.io/mcp/briefboard"
    }
  }
}

If your client only supports stdio servers, bridge the connection with mcp-remote:

{
  "mcpServers": {
    "briefboard": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://app.briefboard.io/mcp/briefboard"]
    }
  }
}

Once authorized, every Briefboard tool becomes available to the agent automatically.

A typical workflow

With the connection in place, the loop is straightforward:

1. Ask the agent to get the next task assigned to it.
2. Have it pull the task prompt — it comes pre-filled with milestone and project context.
3. Let it implement, checking its work against the acceptance criteria.
4. Ask it to log progress in the technical channel and mark the task done.

Because every step reads and writes live Briefboard data, your board stays accurate without anyone updating it by hand.

Keep the human in the loop

The agent can move tasks and leave comments, but the decisions that matter — what to build, when something is truly done, when to merge — stay with you. Briefboard keeps the shared picture honest while the agent does the heavy lifting.

Next, see the full MCP tools guide for everything the agent can do.