MCP

Use discovered documentation tools from the Agno Expert agent.

This walkthrough follows Demo OS at the reviewed revision, which pins Agno 2.8.1. Use the application’s checked-in requirements.

The registered Agno Expert agent (agno-expert) consumes the documentation MCP server through this tool configuration:

from agno.tools.mcp import MCPTools

documentation_tools = MCPTools(url="https://docs.agno.com/mcp")

The application attaches MCPTools to its Agent with persistence, history, and documentation-focused instructions. It discovers the remote tool surface when connecting; the source does not guarantee particular tool names or a tool call on every question.

What MCP gives you

MCP provides tool discovery and calls across a process or HTTP boundary. The Agent’s model chooses among available tools. Connection lifecycle and refresh settings determine when the available tool list is refreshed; a new question does not imply arbitrary rediscovery.

For this pinned SDK, a local command is supplied as a complete command string, such as MCPTools(command="uvx mcp-server-time"), or through supported server parameters. Separate args= and top-level headers= are not accepted constructor options. Static HTTP headers belong in StreamableHTTPClientParams passed as server_params.

When to use MCP vs writing a tool

Use MCP when another process or service owns the capability and its tool contract. Use a Python tool when the application owns the function and can invoke it directly. The documentation server remains an external dependency of this demo.

See it in action

Start Demo OS, choose Agno Expert, and ask how to build an Agno agent. Inspect the discovered documentation tools and their results. A remote server’s current roster and availability must be verified at connection time.

MCP as a context provider in Scout

The standalone Scout application has a separate MCP registry and context-provider configuration. Follow its own source and dependency pin when adapting that pattern.

Next

agents/mcp/agent.py · agents/mcp/instructions.py

Run Demo OS · Current AgentOS documentation