Self-host nexus
0.6.0AI router aggregating MCP servers and LLM providers
- Stars
- 436
- Port
- 8000 (HTTP server (MCP, LLM, health endpoints))
ghcr.io/grafbase/nexusWhat is nexus?
Nexus is a self-hosted AI router that serves as a single gateway for Model Context Protocol servers and large language model providers. It combines tools from local STDIO processes, SSE endpoints, and HTTP-based MCP servers into one endpoint so AI assistants can discover and run tools across all connected servers. For language models, it exposes an OpenAI-compatible API that forwards requests to OpenAI, Anthropic, Google Gemini, and AWS Bedrock with full function-calling support. The software includes OAuth2/JWT authentication, multi-level request and token-based rate limiting with optional Redis storage, fine-grained access control for MCP servers and tools, header rewriting for upstream requests, and OpenTelemetry metrics, traces, and logs. It runs as a single binary or Docker container and is configured through a TOML file with environment-variable substitution.
Should you self-host nexus?
nexus scores 51/100 on selfhost.directory (how scoring works) and ranks #44 of 125 in AI. It has 436 GitHub stars.
There was no stable release in the past 12 months, so check the repository's recent activity before you rely on it.
Setup is moderate: expect some configuration beyond a single command. An official Docker image is available. It is MPL-2.0-licensed and written mostly in Rust.
Its closest alternative on selfhost.directory is ComfyUI (93/100, 143 741 stars). It scores higher overall, so compare the two before you pick. See nexus vs ComfyUI side by side.
Features
- Context-Aware Tool Search
- Docker Ready
- Flexible Configuration
- Header Transformation
- LLM Provider Routing
- LLM Token Rate Limiting
- MCP Access Control
- MCP Server Aggregation
- Model Discovery
- OAuth2 Authentication
- OpenAI Compatible API
- OpenTelemetry observability
- Protocol Support
- Rate Limiting
- Token Forwarding
nexus README
More in API Gateway
nexus-gatewaySelf-hosted open-source AI gateway intercepting LLM traffic at SDK, network, and OS layers with compliance, routing across 20+ providers, caching, quotas, and audit.
NGINXNGINX is a self-hosted, open-source web server that also functions as a reverse proxy, load balancer, content cache, and mail proxy.
OctopusSelf-hosted, open-source LLM API aggregation gateway for individuals: unified channels, protocol conversion, failover, price sync and cost analytics in one binary.
OmniKey AI: Unified Key ManagerOmniKey AI is a self-hosted, open-source LLM proxy routing requests across 12 free providers through unified OpenAI and Gemini endpoints with automatic failover.
OmniRouteSelf-hosted open-source AI gateway routing OpenAI-compatible traffic to 352 providers: 19 routing strategies, quota-aware fallback, 15–95% token compression.
OpenConnectorSelf-hosted open-source gateway that gives AI agents unified access to 1,500+ SaaS providers through 10,000+ prebuilt actions, with MCP, HTTP, and OpenAPI interfaces.
OpenGateLLMSelf-hosted open-source API gateway unifying generative AI models with OpenAI-compatible endpoints, access control, rate limiting, and usage tracking.
OtariA self-hosted, open-source LLM gateway that unifies 40+ model providers behind one OpenAI-compatible endpoint, with virtual keys, budget controls, and usage analytics.
planoOpen-source, self-hosted Envoy-based AI proxy from Katanemo: LLM routing across many providers, multi-agent orchestration, guardrail filter chains and OTel observability.
Qwen GateQwen Gate is a self-hosted, open-source API gateway that lets you use Qwen AI models from chat.qwen.ai as a free, drop-in replacement for the OpenAI API.
RouterlySelf-hosted LLM gateway routing requests across AI providers with multi-policy scoring, cost tracking, budget enforcement, and zero infrastructure dependencies.
rustunnelrustunnel is a self-hosted, open-source tunnel server in Rust exposing local services via encrypted WebSocket. Free managed cloud with pay-as-you-go billing and MCP.
0 comments