# Looker Health MCP Tool

Local ADK agent + MCP Toolbox connector for Looker health checks (`health_pulse`, `health_analyze`, `health_vacuum`).

## Architecture

```mermaid
flowchart LR
  A[ADK Agent] -->|McpToolset| B[MCP Toolbox :5000/mcp]
  B -->|reads| C[tools.yaml]
  B -->|built-in kinds| D[looker-health-pulse / analyze / vacuum]
  D -->|Looker API| E[Looker instance]
```

| Component | Role |
|-----------|------|
| **ADK Agent** | Chat UI (`adk web`) + Ollama model; calls tools via MCP |
| **MCP Toolbox** | Reads `tools.yaml` and exposes built-in Looker health tools over MCP |
| **tools.yaml** | Declares Looker source, tool kinds, and toolset — no custom tool code |
| **Looker API** | Your configured Looker instance |

## Quick start

1. Copy `.toolbox.env.example` → `.toolbox.env` and set Looker API3 credentials.
2. Download the [MCP Toolbox](https://github.com/googleapis/genai-toolbox) binary as `toolbox` (not committed to this repo).
3. Start toolbox: `source .toolbox.env && ./toolbox --tools-file tools.yaml --port 5000`
4. Create venv, install deps: `python3 -m venv .venv && source .venv/bin/activate && pip install -r requirements.txt`
5. Copy env for the agent and run: `adk web --port 8000`

See `looker_health_agent/agent.py` and `tools.yaml` for configuration details.