Connect OpenAI Agents SDK (Python) to Slack, Gmail, GitHub and Notion through a remote Weavz Model Context Protocol (MCP) server. Use the native MCP adapter, then try the app setup recipes with your configured connections.
App setup recipes:
- Slack MCP with OpenAI Agents SDK Python
- Gmail MCP with OpenAI Agents SDK Python
- GitHub MCP with OpenAI Agents SDK Python
- Notion MCP with OpenAI Agents SDK Python
The framework consumes app tools from your configured Weavz MCP server. App recipes require your own app connections; automated checks use synthetic records.
MCP app and framework directory · All examples · Weavz MCP guide
This starter uses the Agents SDK's native MCPServerStreamableHttp adapter. Set up a custom Weavz MCP server in Tool Mode or Code Mode first. The mode is a server-side setting; WEAVZ_MCP_MODE validates the advertised tool names and cannot switch a remote server.
Requires Python 3.11 or 3.12 and uv. Set WEAVZ_MCP_URL to the Streamable HTTP URL and WEAVZ_MCP_TOKEN to its MCP access token. Do not use a Weavz REST API key. The personal https://platform.weavz.io/mcp/weavz server uses OAuth, not this bearer-token flow.
uv sync --locked
uv run python main.py --fixture --mode tools
uv run python main.py --fixture --mode code
uv run pytestThis starts a separate local MCP SDK fixture, calls it through the OpenAI Agents MCP adapter, and shuts it down. The data is synthetic; no model or live Weavz call occurs.
Copy .env.example to .env, set values from your custom server setup, then run uv run --env-file .env python main.py --mode tools or --mode code to discover the server's actual tools through the adapter. No demo alias is assumed on a live server.
For an optional agent run, connect an app containing records for your task, set WEAVZ_MODEL and OPENAI_API_KEY, then run uv run --env-file .env python main.py --mode tools --prompt "Find the latest record matching the task you connected". The server connection stays open through the run and closes afterward.