agentl

Local MCP server that lets an assistant validate, plan, preview and run YAML data pipelines with LLM steps, and debug failed runs.

What you can do

• Validate and plan YAML-defined ETL pipelines before executing them.
• Preview pipeline output and run pipelines end to end, including steps that call an LLM.
• Inspect and debug runs that failed, using a run store kept in SQLite or PostgreSQL.
• Create new pipelines with a configured model and provider.

When to use it

Useful when you build or maintain YAML-based ETL pipelines that mix data transforms with LLM steps, and want an assistant to check definitions, dry-run them, execute them, and investigate failures without leaving the chat.

Requirements

Runs locally over stdio, launched with uvx from the agentl-etl PyPI package, so a Python/uv environment is needed. A run store is configured via AGENTL_DB (SQLite path or postgresql:// URL, default .agentl/agentl.db). Creating pipelines needs AGENTL_MODEL and AGENTL_PROVIDER, plus whatever credentials the chosen LLM provider requires. Desktop only.