Towork

A local-first desktop task manager where AI is a first-class teammate

You and Claude work the same projects, todos, and issues — Claude joins through an embedded MCP server, not a bolted-on chat box. Everything lives in one local SQLite file. No account, no cloud, no telemetry.

Most task managers treat an assistant as a sidecar that talks about your work. Towork makes the AI a teammate that does the work in the same surface you do:

• One shared source of truth. The GUI and the AI mutate the same SQLite database — no sync, no API round-trip, no second copy of your tasks.
• Every change is attributed. Mutations from the GUI are logged as User; mutations over MCP are logged as AI — so the activity log reads like a changelog of who did what.
• The UI reacts to the AI in real time. When Claude creates or updates an item over MCP, the running app live-refreshes (via a towork:changed event) — watch tasks move while the AI works.
• Local-first and private. A single SQLite file on your machine. No account, no cloud, no telemetry.
• Built for the keyboard. A command palette (⌘K), a distraction-free Simple Mode, and a dark "Data Buddy" design system.

Download the latest build for your platform. Each link points at the latest release channel:

Links use the releases/latest/download/ channel. If a direct link 404s after a new version, the filename changed — grab it from the latest release page.

Prefer to build it yourself? See Quick start below.

• Frontend: React 19 + TypeScript + Vite + Tailwind 4, React Router. Dark "Data Buddy" design system (src/styles/).
• Backend: Rust + Tauri 2, rusqlite (bundled SQLite), serde.
• AI integration: an embedded JSON-RPC 2.0 MCP server, over both HTTP and stdio.

Prerequisites: Node 20+ with pnpm, and a stable Rust toolchain (cargo).

pnpm install
pnpm tauri dev      # launches the desktop app (starts Vite automatically)

Other scripts:

pnpm build          # type-check + production web build
pnpm test           # Vitest + React Testing Library suite
pnpm tauri build    # package the desktop app (needs a full icon set; see below)

The SQLite database lives in the OS app-data dir under com.towork.app/towork.db (e.g. ~/Library/Application Support/com.towork.app/towork.db on macOS). It is created and migrated on first launch.

Towork exposes its data to AI clients over the Model Context Protocol (JSON-RPC 2.0), sharing the GUI's database. Mutations made over MCP are attributed to AI in the activity log, and the running app live-refreshes when the AI changes data (via a towork:changed event).

• Tools: list_projects, list_items, create_item, update_item, complete_item, search_items, get_activity
• Resources: towork://projects, towork://project/{id}, towork://item/{id}
• Prompts: daily_review, plan_day

When the desktop app runs, it automatically starts an embedded MCP server on a background thread. Point an HTTP-capable MCP client at it:

http://127.0.0.1:4127/      # POST JSON-RPC; GET returns a health check

Override the bind address with the TOWORK_MCP_ADDR environment variable.

Example client config:

{
  "mcpServers": {
    "towork": { "url": "http://127.0.0.1:4127/" }
  }
}

For clients that spawn their server as a subprocess (e.g. Claude Desktop), the same binary also speaks MCP over stdio:

{
  "mcpServers": {
    "towork": { "command": "/path/to/towork", "args": ["--mcp"] }
  }
}

During development: cd src-tauri && cargo run -- --mcp.