agenttask-mcp

MCP server for AgentTask — lets Claude Code and other MCP clients manage AI agent tasks directly from the conversation.

Setup

1. Get your MCP token

Go to agenttask.io/dashboard/tokens, create a token, and copy the value shown — it's only displayed once.

2. Add to Claude Code

claude mcp add agenttask -s user -- npx -y agenttask-mcp --token=YOUR_TOKEN

That's it. The agenttask server is now available in every Claude Code session.

Verify it's connected:

claude mcp list
# agenttask: ✓ Connected

Available tools

Projects

| Tool | Description | |---|---| | projects_list | List all projects accessible to your token | | projects_get | Get a project by ID | | projects_create | Create a new project | | projects_update | Update a project's name, description, or color | | projects_delete | Delete a project and all its tasks |

Sessions

| Tool | Description | |---|---| | sessions_list | List sessions in a project | | sessions_get | Get a session with its full brief | | sessions_create | Create a session with agent context | | sessions_update | Update a session's name or brief | | sessions_delete | Delete a session (tasks unlinked, not deleted) |

Tasks

| Tool | Description | |---|---| | tasks_list | List tasks with optional filters (project, status, priority, session) | | tasks_get | Fetch a single task by ID or external ID (e.g. PROJ-001) | | tasks_create | Create a new task in a project, optionally in a session | | tasks_update | Update status, priority, assigned agent, or append notes | | tasks_next | Get the highest-priority unblocked task to work on next | | tasks_move | Move a task (and subtasks) to a different project | | tasks_delete | Delete a task | | tasks_add_dependency | Block a task until another completes | | tasks_remove_dependency | Remove a dependency between tasks |

Sessions

Sessions group related tasks within a project (like a sprint or feature set). Each session carries a brief — a markdown document that agents automatically receive when fetching tasks, providing context like architecture decisions or coding standards.

Use session_id when creating tasks or fetching the next task to scope work to a session:

"Create a task in session 5: implement dark mode toggle"
"What's the next task in the auth session?"

Usage in Claude Code

Once connected, Claude can use your tasks naturally:

"What's the next task I should work on?"
"Mark task PROJ-042 as done and note that auth was fixed"
"Create a task: implement dark mode toggle, high priority"
"Move task 42 to project 5"
"List all my projects"
"Create a session called 'Sprint 3' with a brief about the auth refactor"

Configuration

| Flag | Env var | Description | |---|---|---| | --token=VALUE | AGENTTASK_TOKEN | Your MCP token (required) | | --api=URL | — | Override API base URL (default: https://agenttask.io/mcp/v1) |

Using an env var instead of a flag:

claude mcp add agenttask -s user \
  --env AGENTTASK_TOKEN=YOUR_TOKEN \
  -- npx -y agenttask-mcp

Other MCP clients

Cursor / Windsurf — add to your MCP config file:

{
  "mcpServers": {
    "agenttask": {
      "command": "npx",
      "args": ["-y", "agenttask-mcp", "--token=YOUR_TOKEN"]
    }
  }
}

Claude Desktop — add to ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "agenttask": {
      "command": "npx",
      "args": ["-y", "agenttask-mcp", "--token=YOUR_TOKEN"]
    }
  }
}

Links

• agenttask.io — create an account and get your token
• API Reference — full endpoint documentation
• MCP documentation — learn about the Model Context Protocol