Bifrost

The quickest way to start Bifrost is with Docker. The container exposes the gateway and web UI on port 8080.

docker run -p 8080:8080 -v $(pwd)/data:/app/data maximhq/bifrost

If you prefer not to use Docker, you can launch Bifrost directly with npx without a global install step.

npx -y @maximhq/bifrost

Navigate to the web UI at http://localhost:8080. Go to Providers, click Add Provider, select your LLM vendor (e.g., OpenAI), and paste your API key. Bifrost stores keys securely using environment variable references.

# Open: http://localhost:8080

In the web UI under Routing, define a route that maps an incoming model alias (e.g., gpt-4o) to one or more provider-model pairs with priority weights. Enable automatic failover so Bifrost retries a secondary provider if the primary fails or rate-limits.

Bifrost's MCP integration lets AI models routed through the gateway call external tools (filesystem, web search, databases). Enable it in the MCP section of the web UI or via the JSON config, then point your MCP client at Bifrost's endpoint.

Add Bifrost as the AI gateway in your claude_desktop_config.json so all model calls from your client pass through the gateway's load balancer and guardrails.

{
  "mcpServers": {
    "bifrost": {
      "command": "npx",
      "args": ["-y", "@maximhq/bifrost"],
      "env": {
        "BIFROST_PORT": "8080"
      }
    }
  }
}