Skip to main content

AI-Powered StreamWright

StreamWright is designed from the ground up to be AI-first and zero-duplication.

Instead of relying on fragile, out-of-sync markdown documentation or static template repositories, StreamWright makes the connector Python code itself the single source of truth. AI agents, IDE assistants, and local CLI tools dynamically discover connector capabilities, schemas, authentication contracts, and transformation dialect rules at runtime.

The AI Architecture​

1. In-Code Living Contracts​

Every StreamWright connector (such as restapi, postgres, google_ads, s3, meta_ads) exports a standardized 5-part contract directly inside its Python package:

  1. source.yaml specification: configuration requirements, secrets, connection options.
  2. streams/*.yaml specification: request schema, HTTP paths or table queries, pagination parameters.
  3. Authentication & Security: credential schemes, bearer tokens, OAuth2, service accounts.
  4. Transform & Data Shaping: DuckDB SQL capabilities, JSON unnesting, partition joining.
  5. Execution & Behavior: state handling, incremental sync, concurrency, rate limits.

Because these contracts live directly in the code, they never go stale when connectors are upgraded.

2. StreamWright MCP Server (streamwright-mcp)​

Rather than maintaining a separate bespoke AI client or chatbot, StreamWright empowers your host AI agents (Google Antigravity, GitHub Copilot, Claude Code, Cursor, Zed, Windsurf) through the open Model Context Protocol (MCP) standard.

By connecting streamwright-mcp, your host AI assistant gains:

  • Zero-Drift Scaffolding: Inspect live connector schemas to write working source.yaml and streams/*.yaml pipeline definitions.
  • Static DSL Compiler Verification: Run real-time validation against the core StreamWright schema engine directly within the assistant session.
  • DuckDB Dialect Testing: Dry-run transformations and JSON unnesting rules against sample records to guarantee SQL correctness.
  • Up-to-Date Grounding: Directly leverage the latest frontier reasoning models and agent tooling without managing custom LLM clients.

👉 Learn how to configure the MCP server

Getting Started​

To install the Model Context Protocol server in your environment:

pip install streamwright-mcp

If developing from a local clone in an active Python virtual environment:

pip install -e /path/to/streamwright-mcp