Connecting an AI agent (MCP)

Squirlnote runs a built-in MCP server, so an agent like Claude Code can list, create, move, and comment on tasks — the same board you see in the browser, from the agent's side.

1. Create a token

From Settings → Agent Tokens, create a new token. Each token is scoped to your tenant, so an agent connected with it only ever sees your projects.

2. Point your MCP client at Squirlnote

Add Squirlnote as an MCP server in your client's config, using the token from step 1 as a bearer credential. For Claude Code, that's an entry in your MCP settings pointing at your Squirlnote instance's /api/[transport] endpoint.

3. What an agent can do

Once connected, an agent can call tools such as:

• list_projects, list_tasks, list_statuses — read the board.
• create_task, move_task, update_task — act on it.
• add_comment, add_label, add_task_attachment — add context as it works.
• log_accomplishment — record what it finished, for your own review later.

4. The one hard boundary

An agent can move a task through Pending → On-Going → For Review on its own. Moving a task to Done is a human-only action — an agent proposes the work is finished, you confirm it. This keeps "done" meaning what you expect it to mean, even when most of the board activity is agent-driven.