The Peaka MCP server exposes Peaka's data platform to any Model Context Protocol–compatible client. Once connected, your AI assistant can list catalogs, inspect schemas, run SQL, manage caches, save queries, and build semantic catalogs on your behalf — using natural language.
Under the hood it's the same Peaka you use in Studio. The MCP server is a thin protocol adapter that lets agents call Peaka the way humans do.
What you can do with it
- Talk to your data. Ask questions in natural language; the assistant inspects your project's metadata, writes SQL, and runs it on Peaka.
- Manage caches. Create caches, schedule refreshes, and check status without leaving your editor.
- Curate a semantic layer. Save queries, group them under semantic catalogs, and surface them as queryable tables.
- Inspect connections. List data source connections in a project and pull connection-specific configuration.
- Reuse golden SQL. The server queries Peaka's golden-SQL store so the assistant can lean on vetted, hand-written queries before generating new ones.
See Tools reference for the full list of exposed tools.
Three ways to connect
Remote
Point any client that supports remote MCP. Recommended.
NPM
Run the server locally. Authenticates with a Peaka API key.
From source
Clone the repository and run it from a local checkout. For contributors and air-gapped setups.
Authentication at a glance
| Install mode | Auth | Acts as |
|---|---|---|
Remote (mcp.peaka.studio/mcp) | OAuth | The signed-in Peaka user — limited to what they're authorized to do |
| NPM | Project or Partner API key in PEAKA_API_KEY | The owner of the key |
| From source | Project or Partner API key in PEAKA_API_KEY | The owner of the key |
Full details in Authentication.
