voice -agent

VoiceOps: SRE & Kubernetes Agent with MCP/A2A Tools and Skills acts as a conversational AI agent and voice assistant that integrates with MCP servers to perform SRE and Kubernetes operations. The agent uses OpenAI for natural language understanding, Whisper for speech-to-text, ElevenLabs for voice synthesis, and supports A2A (Agent-to-Agent) skills as external tools. You can connect to one or more MCP servers over HTTP/SSE, expose them via an mcp_servers.yaml configuration, and issue natural language prompts that map to available tools across those servers. Typical usage involves configuring your MCP endpoints, starting the agent, and then interacting through voice or text prompts to list, describe, get, or manage resources in your Kubernetes environment or other SRE-related tools that are exposed as MCP capabilities.

Prerequisites:\n- Python 3.9+ (recommended)\n- Access to OpenAI API and an ElevenLabs API key\n- One or more MCP server endpoints (HTTP/SSE) to connect to\n\nStep-by-step installation:\n1) Create and activate a Python virtual environment:\n make venv\n source venv/bin/activate\n\n2) Install dependencies (faster via optional uv):\n make uv # optional, for fast installs\n make install\n\n3) Set required environment variables:\n export OPENAI_API_KEY=your_openai_api_key\n export ELEVEN_API_KEY=your_elevenlabs_api_key\n\n4) Configure MCP servers in mcp_servers.yaml (example provided in README).\n\n5) Run tests (optional):\n make test\n\n6) Run the agent:\n make run\n\n7) Interact with the agent (e.g., say or type a command) to begin.

Tips and notes:\n- The agent currently supports HTTP/SSE MCP servers. If you plan to run stdio-based MCP servers, you may explore Supergateway for exposing them over SSE/WebSockets.\n- Authentication for MCP servers can be configured in mcp_servers.yaml with an auth block using secret_key and env_var. Ensure your env vars are set in your runtime environment.\n- When using A2A integration, you can point to external agents that expose their skills via /.well-known/agent.json and selectively enable tools with allowed_tools.\n- Be mindful of destructive Kubernetes operations. Use allowed_tools filters to limit actions, and test configurations in a safe environment before connecting to production clusters.\n- If you encounter issues with dependencies, validate Python version, virtual environment activation, and that API keys are correctly exported in the shell running the agent.