StreamWright MCP Server

StreamWright MCP is the official Model Context Protocol (MCP) server for StreamWright, the declarative data integration and ELT framework.

It allows AI coding assistants (such as Claude Desktop, Cursor, Google Antigravity, GitHub Copilot, and Zed) to inspect installed connectors, query Draft-7 pipeline schemas, validate YAML sources with pinpoint error locations, and analyze DuckDB SQL transform pipelines in real time — eliminating hallucinations when building data pipelines.

Key Capabilities

• In-Code Connector Contracts: Introspects installed connector packages (restapi, postgres, s3, etc.) directly from Python entry points to extract live contracts, authentication specs, and transport behaviors without static copies.
• Hub Discovery: Discovers uninstalled connectors from the StreamWright Hub registry and provides instant installation recipes.
• Static Pipeline Validation: Validates source.yaml and streams/*.yaml pipelines before execution, returning structured diagnostics with file, line, and column locations.
• SQL Transform Inspection: Parses and traces multi-step DuckDB SQL transformations, table dependencies, and export formats for any stream.
• Core DSL Schemas: Exposes authoritative Draft-7 JSON schemas for editor autocomplete, linting, and LLM grounding.
• CLI Guidance: Steers AI agents to explore live CLI commands (sw run, sw validate, sw connectors, sw mcp) in the local workspace.

Installation

Install via pip:

pip install streamwright-mcp

Or using uv:

uv pip install streamwright-mcp

Configuration

Add streamwright-mcp to your MCP client configuration file (e.g., claude_desktop_config.json, mcp.json, or Antigravity's mcp_config.json).

Standard Configuration

{
  "mcpServers": {
    "streamwright": {
      "command": "streamwright-mcp"
    }
  }
}

Using StreamWright CLI Shorthand (sw)

If you have streamwright installed, you can launch the MCP server directly via sw mcp:

{
  "mcpServers": {
    "streamwright": {
      "command": "sw",
      "args": ["mcp"]
    }
  }
}

Ephemeral Execution (uvx)

Run on-demand without installing into your global environment:

{
  "mcpServers": {
    "streamwright": {
      "command": "uvx",
      "args": ["streamwright-mcp"]
    }
  }
}

Usage

Starting the Server

The server runs over standard input/output (stdio) by default:

streamwright-mcp

You can also specify transport options:

# stdio (default)
streamwright-mcp --transport stdio

# Server-Sent Events (SSE)
streamwright-mcp --transport sse

Or run as a Python module:

python -m streamwright_mcp.cli

Resources

MCP clients that support resource attachments can access:

• streamwright://schema/source — JSON Schema (Draft 7) for source.yaml.
• streamwright://schema/stream — JSON Schema (Draft 7) for streams/<stream>.yaml.
• streamwright://schema/core — Full Markdown specification of the StreamWright DSL.

Documentation & Ecosystem

• Documentation: https://streamwright.web.app
• MCP Guide: https://streamwright.web.app/docs/ai/mcp
• Core Package: https://pypi.org/project/streamwright/
• Repository: https://github.com/karthick-jaganathan/streamwright-mcp