mindbridge

MindBridge is an MCP server that acts as an AI command hub, orchestrating multiple language model providers through a unified OpenAI-compatible API layer. It lets you route prompts to diverse providers (OpenAI, Anthropic, Google, DeepSeek, OpenRouter, Ollama, and OpenAI-compatible services) and supports specialized reasoning models. Core tools include getSecondOpinion to compare responses across models, listProviders to see configured providers and their models, and listReasoningModels for models optimized for complex reasoning. You can configure providers, model defaults, and environment keys via MCP config or environment variables, enabling flexible multi-model workflows and orchestration for agent development, AI backends, and reasoning-heavy tasks. The server can be run in development or production modes, and it exposes an OpenAI-like API surface so existing tools can interact with it with minimal changes.

Prerequisites:\n- Node.js and npm installed on your system\n- Basic familiarity with npm and environment variables\n\nStep 1: Install the MindBridge MCP server globally (recommended)\nbash\nnpm install -g @pinkpixel/mindbridge\n\nStep 2: Run the server (development)\nbash\npm run dev\n\nStep 3: (Optional) Build and run in production\nbash\nnpm run build\nnpm start\n\nStep 4: If you prefer to use MCP config, create or edit an mcp.json file with your server configuration (see README example) and run the MCP runner as usual using your preferred MCP client or IDE. Prerequisites for options include having an API key for the providers you enable (OpenAI, Anthropic, Google, DeepSeek, OpenRouter, etc.).

Tips and considerations:\n- Set environment variables for the providers you intend to use (OPENAI_API_KEY, ANTHROPIC_API_KEY, GOOGLE_API_KEY, DEEPSEEK_API_KEY, OPENROUTER_API_KEY).\n- If you rely on Ollama or an OpenAI-compatible service, ensure base URLs and models align with your local or hosted instances.\n- The MCP config supports provider defaults and default_params like temperature and reasoning_effort to influence model behavior.\n- Use alwaysAllow to expose specific actions (e.g., getSecondOpinion, listProviders, listReasoningModels) to MCP clients without extra authentication.\n- For Smithery deployment, you can install via Smithery CLI as shown in the README, which helps automate client provisioning.\n- If you switch providers or models, update provider_config and default_params to reflect new capabilities and costs.\n- Ensure network accessibility between the MindBridge MCP server and any local model endpoints (e.g., Ollama at http://localhost:11434).