How Do AI Agents Use Bauplan?
AI agents can iterate on code, but not on your data. Bauplan is the execution layer built for fast, AI-generated iteration in production. There are three complementary ways to use AI agents with Bauplan. They can be combined depending on the workflow.
Repository-Based Usage (CLAUDE.md + CLI / SDK)
Bauplan works with LLMs without running an MCP server. In this mode, the assistant operates through repository context: a CLAUDE.md file at the project root that explains how to work with Bauplan, reference documentation for the CLI and Python SDK, and standard command-line execution. This approach is well suited to IDE-based assistants like Claude Code or Cursor and is often the simplest starting point.
Agent Skills for Structured Workflows
Agent Skills are workflow templates that guide LLMs through multi-step tasks. Skills define intent, constraints, and sequencing for creating pipelines, ingesting data, exploring datasets, and diagnosing failures. Skills are distributed as a Claude Code plugin that installs and updates automatically.
MCP Server
The Bauplan MCP Server gives AI assistants direct access to the data lakehouse through the Model Context Protocol. Through the MCP server, an assistant can inspect schemas and tables, run queries, manage data branches and commits, run pipelines, and track jobs. The MCP server is intended for tools like Claude Desktop or the ChatGPT app, where an assistant needs live access to lakehouse state.
A Typical Agent Workflow
- Branch: Create an isolated data branch from production.
- Run: Execute a pipeline (transformation, backfill, schema migration) on the branch.
- Validate: Query the branch to check outputs, row counts, and schema correctness.
- Merge or abandon: If validation passes, merge the branch to publish changes atomically. If validation fails, abandon the branch. Production remains untouched.
Customers like Trust & Will use Bauplan MCP daily alongside other MCP servers to explore data, reason about lineage, and generate and run new pipelines. Moffin uses Bauplan MCP to provide a unified interface for their own customers who interact with their platform in plain English.
AI agents can iterate on code, but not on your data. Bauplan is the execution layer built for fast, AI-generated iteration in production.
