bauplan

The Bauplan MCP Server exposes lakehouse operations to AI assistants and development tooling. It lets Claude Code, Claude Desktop, Cursor, and other agents query Bauplan tables, inspect schemas, manage branches, and execute pipelines against your Bauplan lakehouse. With the MCP server running, you can issue commands or API calls from your assistant to list schemas, preview table metadata, initiate or monitor data pipelines, and fetch results or logs from runs. The server works alongside the repository-based workflow: you can still use CLAUDE.md references and agent skills to guide AI behavior, while the MCP endpoint provides live access to your environment when needed. Typical use cases include validating pipeline definitions against the lakehouse, exploring data lineages, and running WAP-style ingestion pipelines with proper safeguards.

Prerequisites:

• Python installed (recommended 3.11+)
• Access to the Bauplan API (API key)
• uv installed for Python subprocess orchestration (via uvx) or installable tooling per documentation

Installation steps:

1. Clone the repository containing the MCP server (or prepare your environment where you deploy it).
2. Install the MCP server runtime. If using uvx (Python-based MCP runtime), install uvx:

pipx install uvx  # or: python -m pip install uvx

3. Ensure you have a Bauplan API key and set it in your environment or config file. Example (bash):

export BAUPLAN_API_KEY=your-bauplan-api-key

4. Start the MCP server using the MCP package name (as defined in your configuration). If using uvx, the command might resemble:

uvx bauplan-mcp-server

5. Verify the server is listening and accessible from your AI agent or tooling by performing a basic health check or listing available endpoints.

6. If your deployment requires environment-specific adjustments (network, authentication, or CI integration), configure them per your infrastructure guidelines.

Tips and considerations:

• Keep your Bauplan API key secure; do not commit it to source control.
• The MCP server provides live access to lakehouse operations—limit permissions to non-production or sandbox environments when testing prompts.
• If you see authentication errors, verify that BAUPLAN_API_KEY is correctly exported in the environment where the MCP server runs.
• Use the repository-based CLAUDE.md workflow for local development; the MCP server is optional for IDE-based assistants, but required for live operation with AI agents that need lakehouse access.
• When upgrading Bauplan or MCP components, review breaking changes in the API surface and update your skill usage accordingly.