Configuration¶

All configuration lives in backend/.env. Copy the example file to get started:

cp backend/.env.example backend/.env

Environment variables¶

Variable	Default	Description
`LLM_PROVIDER`	`anthropic`	Which LLM to use: `anthropic`, `openai`, or `gemini`
`ANTHROPIC_API_KEY`		Required when `LLM_PROVIDER=anthropic`
`OPENAI_API_KEY`		Required when `LLM_PROVIDER=openai`
`GEMINI_API_KEY`		Required when `LLM_PROVIDER=gemini`
`BEEFREE_MCP_URL`	`http://localhost:8001`	Base URL of the Beefree MCP server
`BEEFREE_CLIENT_ID`		Your Beefree client ID (not used by mock)
`BEEFREE_CLIENT_SECRET`		Your Beefree client secret (not used by mock)
`HOST`	`0.0.0.0`	FastAPI bind address
`PORT`	`8000`	FastAPI port

Switching LLM provider¶

Edit LLM_PROVIDER and fill in the corresponding key. Restart main.py.

AnthropicOpenAIGemini

LLM_PROVIDER=anthropic
ANTHROPIC_API_KEY=sk-ant-...

Model: claude-sonnet-4-5

LLM_PROVIDER=openai
OPENAI_API_KEY=sk-...

Model: gpt-4o

LLM_PROVIDER=gemini
GEMINI_API_KEY=AIza...

Model: gemini-1.5-pro

Nothing else changes. The agent, WebSocket, and frontend are all provider-blind.

Swapping to the real Beefree MCP¶

When you have Beefree SDK credentials:

BEEFREE_MCP_URL=https://your-beefree-mcp.example.com
BEEFREE_CLIENT_ID=your-client-id
BEEFREE_CLIENT_SECRET=your-client-secret

Shut down mcp_mock.py and restart main.py. The agent connects via MCPServerStreamableHTTP(BEEFREE_MCP_URL + "/mcp") - the same MCP Streamable HTTP transport the mock speaks, so the swap is seamless.