Main
- Start with
TOOL_MODE=readoressentialto minimize destructive operations while you validate the integration. - Use a dedicated Metabase API key and rotate it; avoid reusing human admin sessions for automation.
- Log tool calls (dashboard/card/db mutations) so you can audit and roll back unintended changes.
README excerpt (source)
# Metabase MCP Server
[](https://www.npmjs.com/package/@easecloudio/mcp-metabase-server)
[](https://opensource.org/licenses/MIT)
[](https://www.typescriptlang.org/)
[](https://github.com/easecloudio/mcp-metabase-server)
A **Model Context Protocol (MCP) server for Metabase** that gives AI assistants full access to your analytics platform — dashboards, cards, databases, tables, collections, and more.
**Developed and maintained by [EaseCloud](https://easecloud.io)** — cloud-native, AI-driven, and data infrastructure solutions.
## 64 Tools Available
| Domain | Tools |
|---|---|
| Dashboard Management | 16 |
| Card / Question Management | 14 |
| Database Management | 12 |FAQ
Q: How many tools does it expose? A: The README lists tool counts by domain and supports filtering via TOOL_MODE.
Q: Can I run it without global install?
A: Yes—the README recommends npx @easecloudio/mcp-metabase-server.
Q: How do I keep it safe? A: Prefer read-only modes first and use least-privilege API keys for automation.