mcpGW

mcpGW is an MCP server that acts as a gateway and proxy designed to simplify how applications connect with and orchestrate multiple tools. Derived from the mcphost project, it allows developers and systems engineers to bridge local inference providers, voice interfaces, and custom Model Context Protocol servers under a unified endpoint. The tool connects primarily with Ollama for local LLM inference and supports modular integrations such as Whisper.cpp for speech-to-text and MeloTTS for text-to-speech processing. By managing external MCP servers, such as reminder services or custom background utilities, mcpGW handles the execution lifecycle and communication layer between downstream clients and upstream tools. It serves developers looking to run self-hosted AI agent architectures on local hardware without depending on external cloud providers. The gateway exposes configurable web APIs and includes a built-in user interface server to streamline the testing and deployment of integrated toolsets.

Category: AI & LLM Tooling

Tags: gateway, infrastructure, orchestration, proxy

Visit mcpGW

How to install and configure mcpGW

1. Ensure prerequisites are installed and running: set up Ollama with your chosen model (such as ollama pull mistral and ollama serve). Optionally run Whisper.cpp server and MeloTTS if voice capabilities are needed. 2. Clone the repository and navigate into the project directory. 3. Create your configuration file based on config.sample.json to define your UI listener, speech services, inference endpoints, and connected mcpServers. 4. Compile the application binary by running: bash make all 5. Start the proxy server with your configuration file: bash ./mcpgateway --config ./config.sample.json 6. For further client integration details, refer to the project repository at https://github.com/thirdmartini/mcpgw.

What you can do with mcpGW

• Serving as a central MCP proxy to route tool requests between Ollama models and multiple local subprocesses. - Hosting custom web application user interfaces served directly through the built-in HTTP server runtime. - Integrating Whisper.cpp to enable speech-to-text prompt transcription for local Model Context Protocol tools. - Synthesizing tool responses into spoken voice using the optional local MeloTTS text-to-speech engine. - Orchestrating custom child MCP servers like reminder managers through declarative JSON execution configurations.

Key facts

• https://github.com/thirdmartini/mcpgw
• AI & LLM Tooling
• gateway, infrastructure, orchestration, proxy

Part of MCP Servers

What is mcpGW?

mcpGW is an open-source Model Context Protocol gateway and proxy derived from mcphost. It coordinates multiple MCP tools and integrates local LLM execution, optional speech recognition via Whisper.cpp, and speech synthesis through MeloTTS.

Which inference backends does mcpGW support?

The project is currently tested and configured for local inference using Ollama. You can point the inference provider to your local Ollama host, port, and preferred language model.

How do I add custom MCP tools to mcpGW?

You can register custom MCP servers by declaring them in the config.sample.json file under the mcpServers block, specifying the executable command and any required runtime arguments.

Does mcpGW require speech processing services?

No, speech-to-text and text-to-speech services are optional. You can configure mcpGW purely for text-based inference and MCP server orchestration by leaving the speech blocks empty or omitted.