MUXI is a framework for building and orchestrating multi-agent systems with persistent memory and standardized MCP (Model Context Protocol) integration. The server exposes agents as MCP-compatible endpoints and supports multiple transport modalities (HTTP+SSE, REST, WebSockets, and WebRTC) to enable real-time, streaming, and multi-modal interactions between agents and external clients. Use the MCP interface to discover agent capabilities, register tasks, route messages via the intelligent router, and manage memory contextualization across agents. The built-in tools enable you to discover and invoke agent capabilities, stream results via SSE, and maintain secure authenticated access across REST APIs and MCP endpoints. With its modular architecture, you can enable only the components you need (e.g., memory backends, MCP bridge, or UI) and extend it with additional agents or memory systems as your use case grows.

To get started, run the MCP server (Muxi) and connect a client via the MCP transport of your choice. You can use the SSE-based MCP host integration for streaming responses, or a CLI/REST client to manage configuration and perform standard requests. The server will expose agent capabilities, allow task delegation, and provide context memory management so agents can share information while preserving isolation where needed. Explore the capabilities by discovering registered agents, invoking tasks, and observing how messages are routed to appropriate agents based on capabilities and context.

Prerequisites:

• Node.js (LTS version) and npm installed on your machine
• Basic familiarity with MCP concepts and how to interact with REST/SSE/WebSocket endpoints

Install steps:

1. Clone the repository or download the release package: git clone https://github.com/ranaroussi/muxi.git cd muxi

2. Install dependencies for the server component: npm install

3. Build or prepare the server entry point (if required by the project structure):

   If a build step is provided, run it. Otherwise, ensure server.js is available at the project root.

   npm run build || echo "No build script defined; proceeding with runtime server.js"

4. Run the MCP server locally: npm run start || node server.js

5. Verify the server is up by checking the MCP endpoint (e.g., http://localhost:PORT or the SSE/WebSocket endpoints described in the docs). If your configuration uses a different port, adjust accordingly.

Notes:

• If you plan to run behind a reverse proxy or in production, configure TLS, authentication, and proper memory backend connections as described in the docs.
• Ensure environment-specific variables (e.g., memory backends, database URLs) are set in your environment or in a dedicated .env file.

Tips and common considerations:

• Memory backends: MUXI supports multiple memory systems (FAISS, sqlite-vec, pgvector, Memobase). Ensure the chosen backend is configured and accessible (e.g., database URLs, file paths).
• MCP compatibility: Use the MCP server endpoint with SSE for streaming results or a bridge for non-SSE clients. Leverage task delegation and capability discovery to orchestrate agent interactions.
• Security: Use authentication shared with the REST API and consider implementing access controls for MCP endpoints to prevent unauthorized agent interactions.
• Declarative configuration: Define agents via YAML or JSON files to minimize code changes and simplify reconfiguration.
• Extensibility: The framework is modular; enable only the components you need (CLI, REST API, Web UI, or memory modules) to reduce resource usage.
• Debugging: If you encounter connection or capability discovery issues, check the MCP transport settings, ensure proper port exposure, and review logs for errors related to memory backends or agent registration.