The connector-builder-mcp
repository provides a Model Context Protocol (MCP) server implementation for Airbyte connector building operations. This repository will eventually codify relevant parts of the builder-ai
functionality, with a focus on AI ownership rather than AI assist.
This project emphasizes end-to-end AI ownership, including:
- Autonomous connector building and testing
- Automated PR creation and management
- Complete workflow automation without human intervention
This differs from AI assist tools that merely help human developers - instead, this enables AI agents to fully own the connector development process from start to finish.
The MCP server follows the established PyAirbyte pattern with:
- Main server module that initializes FastMCP
- Separate tool modules for different functional areas
- Comprehensive connector building capabilities exposed as MCP tools
- Manifest Operations: Validate and resolve connector manifests
- Stream Testing: Test connector stream reading capabilities
- Configuration Management: Validate connector configurations
- Test Execution: Run connector tests with proper limits and constraints
To use the Builder MCP server, you'll need Python 3.10+ and uv for package management.
For detailed development setup and contribution guidelines, see CONTRIBUTING.md.
uv sync --all-extras
Start the MCP server:
# You can use any of these to start the server manually:
uv run connector-builder-mcp
poe mcp-serve-local
poe mcp-serve-http
poe mcp-serve-sse
Or use with MCP clients by configuring the server in your MCP client configuration.
To use the Builder MCP server with MCP clients like Claude Desktop, add the following configuration:
{
"mcpServers": {
"connector-builder-mcp": {
"command": "uvx",
"args": [
"connector-builder-mcp"
]
}
}
}
{
"mcpServers": {
"connector-builder-mcp": {
"command": "uvx",
"args": [
"--from",
"git+https://github.com/airbytehq/connector-builder-mcp.git@main",
"connector-builder-mcp"
]
}
}
}
You can run from a locally cloned version of the repo using the below syntax.
Remember to update the path to your actual repo location.
{
"mcpServers": {
"connector-builder-mcp": {
"command": "uv",
"args": [
"run",
"--project",
"/path/to/repos/connector-builder-mcp",
"connector-builder-mcp"
]
}
}
}
For convenience, you can use Poe the Poet task runner:
# Install Poe
uv tool install poethepoet
# Then use ergonomic commands
poe install # Install dependencies
poe check # Run all checks (lint + typecheck + test)
poe test # Run tests
poe mcp-serve-local # Serve locally
poe mcp-serve-http # Serve over HTTP
poe mcp-serve-sse # Serve over SSE
You can also run poe --help
to see a full list of available Poe commands.
If you ever want to see what a Poe task is doing (such as to run it directly or customize how it runs), check out the poe_tasks.toml
file at the root of the repo.
This project uses uv for package management and follows modern Python development practices.
# Install dependencies
uv sync --all-extras
# Run linting
uv run ruff check .
# Run formatting
uv run ruff format .
# Run tests
uv run pytest tests/ -v
# Type checking
uv run mypy connector_builder_mcp
Helping robots build Airbyte connectors.
For comprehensive testing instructions, including FastMCP CLI tools and integration testing patterns, see the Testing Guide.
See the Contributing Guide for information on how to contribute.