Model Context Protocol

Bring your tasks into
any AI assistant.

The companion Model Context Protocol server for Tasks Manager. It reads your projects and tasks straight from the app's local database — so Claude and other MCP hosts can find, read, create and update tasks for you, with an optional native approval prompt before every write.

Latest npm version·Node 20 or later

Needs the desktop app. The server reads Tasks Manager's local database, so the app must be installed and running on the same machine.

Get the app

Set up in a minute

Install, connect, done.

1

Install the server globally

Terminal
npm i -g tasks-manager-mcp
2

Register it with your MCP host

For Claude Code, add it straight from the CLI:

Terminal
claude mcp add --scope user tasks-manager -- tasks-manager-mcp

Or add it to any host's config (e.g. claude_desktop_config.json):

mcp config
{
  "mcpServers": {
    "tasks-manager": {
      "command": "tasks-manager-mcp"
    }
  }
}
3

Ask your assistant

With the app running, ask Claude to find a task, summarize a project, or capture a finding — it works directly against your real tasks, no copy-paste.

Optional

Let an agent take over — no per-write prompts

By default every create_task, update_task and add_task_findingpops a native desktop dialog and waits for you to confirm — and while it's on, a prompt alone can't skip it. It's opt-out, though: to let an agent create and update tasks autonomously, set TASKS_MANAGER_MCP_AUTO_CONFIRM=1 in the server's env.

mcp config · auto-confirm
{
  "mcpServers": {
    "tasks-manager": {
      "command": "tasks-manager-mcp",
      "env": { "TASKS_MANAGER_MCP_AUTO_CONFIRM": "1" }
    }
  }
}

With this set, writes apply silently and the agent has full write access to every unlocked project — only do this on a machine you trust. Locked projects stay invisible and unwritable regardless. To restore the prompts, remove the variable (or set it to anything other than 1).

What it can do

Eight tools, read and write.

Read

  • list_spaces

    Every workspace, with accessible vs. locked project counts.

  • list_projects

    Browse projects, optionally scoped to a single space.

  • find_task

    Fuzzy find by code (OCOI-4), id, number or title substring.

  • get_task

    Full task detail in one call — description, findings and referenced images as native blocks.

  • list_tasks

    Browse tasks filtered by project, status or tag.

Write

  • create_task

    Create a task in any project — status, priority, tags, subtasks and more.

  • update_task

    Partial update by code or id, with auto section re-mapping on status change.

  • add_task_finding

    Append a markdown finding to capture notes beside the work.

Safe by default

You stay in control.

Approve writes before they happen

By default, before any task is created or updated the server pops a native desktop dialog showing the exact change — nothing is written until you confirm. Opt into hands-free automation when you want it.

Locked projects stay invisible

Password-locked projects can't be enumerated, read or written by any tool. Unlock a project in the app to expose it to MCP.

Runs entirely on your machine

It opens the app's local database directly on the same host — no cloud, no network round-trips, nothing uploaded anywhere.