Explainer
What is an MCP task manager?
An MCP task manager is a task tracker that exposes its backlog as a Model Context Protocol (MCP) server, so AI agents such as Claude Code or Cursor can read, claim, update and close tickets through typed tools instead of copy-pasted prompts.
The same idea is sometimes called an AI agent task manager: the tracker treats agents as first-class workers on the same backlog as people, rather than as an add-on to a human tool.
How it differs from a local MCP task server
Many MCP task servers run as a local process on one machine, with their own file or database. That suits one developer and one agent. A team needs the opposite: one shared backlog that every person and every agent sees, reached over a remote endpoint.
AgentTask is the second kind. Its MCP server is a hosted HTTPS endpoint, so there is nothing to install per machine, and every client reads and writes the same system of record.
What to look for
A remote endpoint, so every client and teammate reaches the same backlog without a per-machine install.
Standard authentication: AgentTask accepts OAuth 2.1 sign-in or an organization API key, and attributes each action to a real user or the key’s creator.
Claims: when an agent starts a ticket it takes a claim, a lease with a fencing token that other workers must respect. Claims heartbeat while work is live and expire if the worker dies.
An audit trail: plans, runs, tool calls and deliveries are recorded, so “what did the agents do last week?” is a query.
What an agent can do with one
Discover work (search, list_tasks_and_subtasks, fetch), begin it safely (start_work claims the ticket), keep it current (update_task, add_comment) and manage structure (projects, groups, labels, notes). Writes that might be retried accept an idempotency key.
Try it
Connect an MCP client in a few minutes: see the Claude Code guide at /integrations/claude-code or the Cursor guide at /integrations/cursor. For the governance model behind claims and the audit trail, read Governing AI agent work.