Kong/kong
Kong Gateway is a cloud-native, scalable API gateway that manages API, AI LLM, and Model Context Protocol (MCP) traffic with advanced security, observability, and extensibility features.
teabranch/open-responses-server
open-responses-server is a middleware server that exposes the OpenAI Responses API in front of whatever model backend the operator already runs, such as Ollama, vLLM, LiteLLM, Groq or OpenAI itself. Its purpose is to let clients that require the Responses API, notably the OpenAI coding assistant Codex, work against self-hosted models, translating between the Chat Completions interface a backend speaks and the Responses interface the client expects while keeping conversation state and tool calls working. Its relevance to the Model Context Protocol is that MCP servers are a first-class part of the tool layer: the server reads a JSON file of MCP servers given by MCP_SERVERS_CONFIG_PATH, keeps a cache of the tools those servers advertise, refreshed on the interval set by MCP_TOOL_REFRESH_INTERVAL, and makes them available around both the Chat Completions and the Responses endpoints, so a plain OpenAI API client gains MCP tools without implementing the protocol itself. Configuration is entirely by environment variable: OPENAI_BASE_URL_INTERNAL points at the backend, OPENAI_BASE_URL is the address of this server, OPENAI_API_KEY is tunnelled to the backend, API_ADAPTER_HOST and API_ADAPTER_PORT control binding, STREAM_TIMEOUT and HEARTBEAT_INTERVAL govern streaming and server-sent event keepalive, MAX_CONVERSATION_HISTORY and MAX_TOOL_CALL_ITERATIONS bound stored turns and tool loops, and LOG_LEVEL with LOG_FILE_PATH control logging. Installation is from PyPI as open-responses-server, from source with uv, or from the published multi-architecture container image; the otc command line tool starts the server and offers an interactive configure step. The README marks tool calls, MCP support, tests, Docker images, the PyPI release and CLI validation as done, while persistent rather than in-memory state, web search, file upload and search, a code interpreter and computer use APIs are listed as planned.
https://github.com/teabranch/open-responses-server
Also in Gateways
alibaba/higress
Higress is a cloud-native AI native API gateway by Alibaba that supports hosting Model Context Protocol (MCP) servers, enabling unified management of AI and LLM APIs with extensible Wasm plugins and enterprise-grade features.
IBM/mcp-context-forge
IBM/mcp-context-forge is a production-grade MCP Gateway that unifies and manages MCP and REST services, providing a secure, scalable, and feature-rich interface for AI clients.
open-webui/mcpo
mcpo is a secure and simple proxy server that exposes Model Context Protocol (MCP) tools as OpenAPI-compatible HTTP servers, enabling seamless integration with LLM agents and applications.
sparfenyuk/mcp-proxy
mcp-proxy is a proxy server tool that enables communication between MCP servers using stdio and SSE transport protocols, facilitating interoperability within the MCP ecosystem.
metatool-ai/metamcp
MetaMCP is a proxy server that aggregates multiple MCP servers into a unified MCP server endpoint, applies middleware, and hosts these endpoints for seamless integration with MCP clients.
samanhappy/mcphub
MCPHub is a unified management platform that aggregates multiple Model Context Protocol (MCP) servers into streamable HTTP endpoints, providing centralized control, real-time monitoring, secure access, and intelligent tool discovery for efficient AI tool integration.
mcp-ecosystem/mcp-gateway
MCP Gateway is a lightweight, configuration-driven gateway service that converts existing APIs into MCP protocol-compliant endpoints without code changes, supporting Docker and Kubernetes deployments with a built-in management UI.