boardagent-mcp
# BoardAgent
A task manager that lives on your computer — for you **and** your AI agents.
Terminal UI, REST API, MCP server. Free, local, no cloud.
## Install (Windows)
**You need the prebuilt exes.** Two ways to get them:
**Option A — use the release binaries (easiest):**
Download the latest release from GitHub → unzip → double-click `INSTALL.bat`.
That's it. It copies the programs, sets up autostart, installs a watchdog,
and adds the app to your PATH.
**Option B — build them yourself:**
Requires Python 3.10+.
```bash
pip install -e . # installs the 3 CLI commands
python scripts/build_exes.py # builds boardagent.exe, boardagent-server.exe, boardagent-mcp.exe into dist/
```
Then run `INSTALL.bat`.
## Run
**Windows (after INSTALL.bat):** double-click `boardagent.exe` (or type
`boardagent` in a terminal). The background server auto-starts at logon.
**Anywhere (source):**
```bash
boardagent-server # terminal 1 — the background service
boardagent # terminal 2 — the task board UI
```
First-time tips:
- Press `c` to create a task, `a` for AI mode, `q` to quit.
- Full keyboard-first: arrows move, space selects/opens, enter activates.
- Settings tab: theme, opacity, API keys, keybinds.
## Let an AI agent use the board
Point your MCP host (Claude Desktop, Cursor, Hermes) at `boardagent-mcp`:
```json
{
"mcpServers": {
"boardagent": { "command": "C:\\path\\to\\boardagent-mcp.exe" }
}
}
```
Create an API key in Settings → API Keys for anything else.
## How it works
- **REST API** — source of truth, `http://127.0.0.1:7373`.
- **MCP server** — thin adapter over the same service layer.
- **Textual TUI** — the app you see, talks to the service over localhost.
- Data lives in `~/.boardagent/boardagent.db` (SQLite). Everything local.
## Docs
- **For humans**: `docs/human/` — intro, MCP setup, themes, packaging.
- **For AI agents**: `docs/agent/` — token-optimized REST + MCP references.
## Development
```bash
python -m pytest
python scripts/generate_agent_docs.py # needs the server running
python scripts/build_exes.py # Windows exes into dist/
```
## License
MIT — see `LICENSE`.
TDQS
Scored across 7 tools
Each tool maps to a distinct action: CRUD (create/list/get/update/delete) plus two clearly-scoped lifecycle operations (claim for locking, complete for finishing). Claim vs update are well separated by their descriptions (lock vs metadata merge), so misselection is unlikely.
All seven tools share the boardagent_ prefix and follow a strict verb_noun pattern (create_task, list_tasks, get_task, delete_task, update_task, claim_task, complete_task). No mixed conventions or vague verbs.
Seven tools is a well-scoped set for a task board, covering the essential operations without bloat. Every tool earns its place.
Full CRUD plus claim/complete lifecycle is strong, but a few natural operations are absent: releasing/unclaiming a task, reopening a completed task, and explicit assignment. These are minor gaps an agent can largely work around.