Alkemi MCP

What alkemiai/alkemi-mcp MCP server does

alkemiai/alkemi-mcp MCP server exposes Alkemi data products to MCP-compatible clients through a local STDIO process. The connected Alkemi environment can use data from Snowflake, Google BigQuery, Databricks, and other sources. Rather than requiring each agent to carry its own database schema notes and query examples, Alkemi keeps that context with the data product and uses it to generate and execute database queries.

This setup is intended for teams that want multiple MCP clients or agents to query the same governed data context. A product can be selected explicitly when the client should only work with one part of the available Alkemi data.

How it works

The package acts as a STDIO wrapper around Alkemi’s Streamable HTTP MCP endpoint:

https://api.alkemi.cloud/mcp

An MCP client starts the local process and communicates with it over standard input and output. The wrapper sends requests to Alkemi using the configured bearer token. Alkemi supplies the stored database metadata and query guidance, creates an appropriate query, and queries the connected source.

The PRODUCT_ID setting optionally narrows the server to one Alkemi product. Multiple instances can be configured with different product IDs, allowing separate client entries for different datasets. MCP_NAME can provide distinct names when more than one instance is registered in the same client.

Setup and configuration

Create an API key through datalab.alkemi.ai, then provide it as BEARER_TOKEN. The README shows installation through the npm package @alkemiai/alkemi-mcp and configuration for Claude Desktop.

A direct package-based configuration uses npx to launch the server. BEARER_TOKEN is required for the STDIO integration. MCP_NAME is optional unless multiple instances would otherwise share the same client name. PRODUCT_ID is optional and accepts the ID of the product to scope.

For local development, the repository can be installed with npm install, built with npm run build, and run from the generated build/index.js. npm run watch enables automatic rebuilding during development. The repository also provides npm run inspector for MCP debugging through the MCP Inspector.

Tools and capabilities

The material describes these capabilities:

• Query Alkemi-connected Snowflake data.
• Query connected Google BigQuery data.
• Query connected Databricks data.
• Work with other data sources connected through Alkemi.
• Use stored metadata and query examples to inform generated queries.
• Scope an MCP instance to a selected Alkemi product.

The exact individual MCP tool names are not listed in the provided material, so this description does not infer them.

Limitations and notes

alkemiai/alkemi-mcp MCP server depends on an Alkemi account and a bearer token, as well as data products already connected to Alkemi. It does not directly document database credentials, client support beyond the shown Claude Desktop setup, or the full list of supported source systems. Query quality depends on the metadata and examples maintained for the relevant data product.

The provided Claude Desktop examples include placeholder paths for locally built files; those paths must be replaced with an actual checkout location when using the repository build instead of the npm package.