# bifrost [Health: Active]

**Category:** 🔗 Aggregators  
**Repository:** https://github.com/maximhq/bifrost  
**GitHub Stars:** 9379  
**Views:** 0  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/bifrost

## Description
Open-source, high-performance AI gateway that works as your LLM gateway, MCP gateway and agent gateway. One OpenAI-compatible API for 10,000+ models, with failover, load balancing, semantic caching, virtual keys, budgets and MCP Code Mode. Works with any SDK or agent and adds under 200µs of overhead at 5k RPS.

## Claude Desktop Quick Installation
No confirmed install details yet, so no config is published. Follow the setup steps in the repository above.

## Documentation

## What bifrost does

bifrost combines model routing and MCP gateway functions in a single AI infrastructure layer. Applications can send requests through one OpenAI-compatible API while bifrost manages connections to more than 23 model providers, including OpenAI, Anthropic, AWS Bedrock, Google Vertex, Azure, Cerebras, Cohere, Mistral, Ollama, and Groq.

Its MCP capability allows AI models to use external tools, with the project identifying filesystem operations, web search, and databases as examples. The wider gateway also covers request reliability, cost controls, access management, and operational visibility.

## How it works

The gateway accepts HTTP requests using familiar OpenAI-compatible routes, such as `/v1/chat/completions`, and uses the requested model identifier to route traffic to a configured provider. Multiple providers or API keys can be used for automatic fallback and load distribution. Semantic caching can reuse responses for semantically similar requests, reducing repeated work and latency.

A built-in web interface provides configuration, monitoring, and analytics. Configuration can also be managed through the API or files. For Go applications, the project offers a core SDK that can be embedded directly instead of running the HTTP gateway as a separate service.

MCP gateway support is described as an enterprise capability. This makes bifrost relevant to deployments that need centralized control over model access and external tool use, rather than only a local, standalone MCP process.

## Setup and configuration

The gateway can be started locally with NPX:

```bash
npx -y @maximhq/bifrost
```

Docker is also supported:

```bash
docker run -p 8080:8080 maximhq/bifrost
```

After startup, open `http://localhost:8080` to use the web interface. The gateway can then be called over HTTP on port 8080. A persistent data directory can be mounted for Docker deployments:

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

Provider configuration is available through the web UI, API, or configuration files. Secrets can be supplied through environment variables and deployment secrets.

## Tools and capabilities

- Route requests through a common OpenAI-compatible API.
- Connect applications to multiple model providers.
- Fail over between providers and models.
- Balance traffic across providers and API keys.
- Let models use external MCP tools such as filesystem, web search, and database tools.
- Apply semantic caching, budgets, rate limits, and virtual-key access controls.
- Expose Prometheus metrics, distributed tracing, logging, and analytics.

## Limitations and notes

The project presents MCP gateway functionality as part of its enterprise deployment capabilities. The documented quick start focuses on the general HTTP gateway and provider routing, not on a standalone local MCP server configuration. Provider availability, governance features, and MCP gateway access may therefore depend on the deployment edition and configuration.

_Full upstream README: https://allmcps.com/mcp/bifrost/readme_