native-mcp

To connect an agent to MCP servers and expose their tools automatically, you need a configured mcp_servers block in your ~/.hermes/config.yaml file and the mcp Python package installed. This enables startup discovery and automatic tool injection into agent toolsets.

The best way to integrate stdio and HTTP MCP servers with an agent is using a tool that supports both protocols for flexible deployment. This auto-discovers available tools at startup and injects them transparently into agent toolsets for seamless access.

No, you do not need to manually register tools when adding MCP servers. The system automatically discovers available MCP tools on startup and transparently injects them into all agent toolsets, exposing them as first-class tools across conversations.

Yes, you can use Python to manage MCP server integration by installing the mcp Python package. Combined with a configured mcp_servers section in your config.yaml, it enables automatic discovery and injection of external tools into agent workflows.

Tool discovery works by automatically scanning configured MCP servers at startup. It identifies available tools across both stdio and HTTP servers and transparently injects them into all platform toolsets, allowing seamless cross-server access without manual bridging.

If MCP tools are not showing up in the agent after startup, ensure the mcp Python package is installed and your ~/.hermes/config.yaml contains a properly configured mcp_servers block. These two elements are required to trigger startup discovery and automatic tool injection.