Skip to main content
MCP is a protocol that standardizes how an agent discovers tools, requests capabilities, and exchanges structured data with external systems.

​
Key concepts

  • Host: runs the LLM/agent
  • Server: exposes a catalog of tools with schemas and permissions
  • Client: connects the agent to the server
  • Tools: typed functions the agent can call (e.g., search, file IO, API)

​
Why MCP

  • Unified discovery of available tools
  • Strong typing and metadata for safer calls
  • Isolation and permission prompts for sensitive actions

​
Example capability descriptor

{
  "tool": "search",
  "description": "Query indexed docs",
  "inputSchema": { "type": "object", "properties": { "q": { "type": "string" } }, "required": ["q"] },
  "rateLimit": { "rpm": 60 }
}

​
Learn more

  • Intro to agents → /learn/ai-agent
  • Build one end‑to‑end → /learn/build-agents