- CollectionLLM Gateway & Routing
- Classificationagent-native
- LicenseApache-2.0
- MCPAvailable
- Agent SkillUnavailable
- InterfacesMCP · CLI · SDK · json rpc · http · API
- Verified2026-10-13
What it does
Bifrost is a high-performance, self-hosted AI gateway that provides one OpenAI-compatible interface to more than 23 model providers. It handles provider configuration, request routing, retries, automatic fallbacks, load balancing, semantic caching, and drop-in integrations for common AI SDKs.
For agent systems, Bifrost adds an MCP client and gateway, opt-in Agent Mode for automatic tool execution, per-virtual-key tool filtering, hierarchical budgets and rate limits, and request observability. The gateway can aggregate tools from multiple MCP servers behind one endpoint while keeping provider credentials and governance policies outside agent code.
Agent-facing interfaces
- Catalog protocols: MCP, CLI, SDK, json rpc, http, API
- MCP: ✅ Available
- Agent Skill: ⚠️ No official Agent Skill published by Bifrost yet.
- Onboarding type:
cli
How agents onboard
npx -y @maximhq/bifrost
See the full onboarding section in the dossier.
Protocol surface
| Interface | Detail |
|---|---|
| OpenAI-compatible HTTP API | Unified /v1 API for supported model providers, including chat, responses, streaming, and tool calling where supported. |
| MCP Gateway | POST /mcp JSON-RPC tool discovery/execution and GET /mcp SSE for external MCP clients. |
| Go SDK | go get github.com/maximhq/bifrost/core for direct in-process provider access and tool-calling integrations. |
| Configuration API | Provider, virtual-key, budget, rate-limit, and MCP configuration through the management API. |
| Configuration files | Declarative config.json configuration for provider and governance setup. |
| Observability | Request logs, cost/token/latency metadata, Prometheus metrics, and distributed tracing integrations. |
Verification
Last verified 2026-10-13.
Sources:
- https://github.com/maximhq/bifrost