Skip to main content

Using StreamWright with AI assistants

StreamWright lets AI coding assistants write sources and jobs from what is actually installed, not from memory. Each connector describes itself in its own Python package. The MCP server hands that description to the assistant, together with the schemas, guides and checks StreamWright itself uses.

In short

Install streamwright-mcp and add it to your assistant. The assistant then reads the contracts of your installed connectors and checks the YAML it writes, before you run anything.

How it works​

Contracts live in the connector code​

Every connector (for example restapi, postgres, google_ads, s3, meta_ads) declares its contract inside its Python package:

  • its connection and authentication keys, and its secrets;
  • its operations and query builders;
  • what each engine supports.

The contract ships with the code, so it changes when you upgrade the connector. The assistant always sees the version you have installed.

The MCP server gives assistants that knowledge​

streamwright-mcp is a Model Context Protocol server. It works with the assistant you already use (GitHub Copilot, Claude Code, Cursor, Google Antigravity, Zed, Windsurf and others); StreamWright has no chatbot of its own. Through it the assistant can:

  • read connector contracts and guides, engine guides, JSON Schemas and CLI help;
  • search all of those guides and the CLI help by keyword;
  • validate source.yaml, streams/*.yaml and job files with the same checks as streamwright validate;
  • inspect how a stream is planned: its requests, steps, tables and columns.

The server never runs a source or a job, and never reads secret values.

Get started​

pip install streamwright-mcp

Then connect it to your assistant.

See also​

  • MCP server: setup per assistant, and the full tool list.
  • CLI reference: streamwright validate runs the same checks from a terminal.