drunk-mcp-proxy copilot-instructions.md

Copilot Instructions for Drunk MCP Proxy

Quick Orientation

Drunk MCP Proxy is a Python 3.10+ FastAPI/Starlette gateway that fronts multiple MCP (Model Context Protocol) servers and LLM providers. The app composition is config-driven—configuration loaded from data/config.yaml (not JSON), then providers are instantiated and mounted as sub-applications.

Key entry point: src/main.py → src/app/starlette_app.py → provider instances mounted via Starlette.

Essential Patterns & Conventions

Project Structure & Imports

• Layout: src/ based, with src/app/, src/auth_providers/, src/proxies/, src/tools/, tests/.
• Imports: Use relative imports from src/ root (no src. prefix), e.g. from app.config_provider import ConfigProvider, from tools.env import SERVER_NAME.
• When patching in tests: Match import path as used in code, e.g. @patch("src.proxies.llm_proxies_provider.AppConfigProvider.get_instance").

Configuration & Environment

• Config file: data/config.yaml with three root sections: auth, llm, mcp.
• Field naming: Use snake_case in YAML, NOT camelCase (e.g., default_provider, mcp_servers).
• Config loading: AppConfigProvider.get_instance() loads once at startup; use .get_*_config() methods to retrieve typed sections.
• Environment variables: Checked via src/tools/env.py; override via OS env (e.g., FASTMCP_AUTH_ENABLED, MCP_OAUTH_STORAGE_TYPE).

Testing & Mocking

• Run tests: python -m pytest (adds current directory and src/ to sys.path via conftest.py).
• Mocking environment: Use monkeypatch.setattr() to override module-level variables; mocking both regular and async functions.
• Mocking FastAPI apps: Use @patch() on provider methods like _get_openai_client() or _get_fastapi_app() to inject fake clients/apps; use TestClient from starlette.testclient.
• Request mocking: Create Mock() objects with .headers, .json(), .form() as needed; monkeypatch provider methods that return client instances.