MCP (Model Context Protocol)
An open standard for connecting AI applications to external tools and data sources.
Simple explanation
MCP is a protocol that lets any AI agent use any tool server. Write a tool server once, any MCP-compatible agent can use it. This decouples tools from agent frameworks.
Engineering explanation
MCP defines a client-server protocol: an MCP server exposes tools (functions), resources (data) and prompts (templates) via JSON-RPC over stdio or HTTP. An MCP client (your agent) connects, discovers capabilities and calls them. This eliminates tool duplication across frameworks — write a Postgres MCP server once, use it with LangGraph, Claude, or any MCP client.
Example
MCP server exposes query_postgres(sql) → any MCP-compatible agent can call it → no framework-specific tool code.
When to use it
When you have multiple agent frameworks or want reusable tool ecosystems. Overkill for a single agent with 2-3 tools.
Common misconception
MCP does NOT make agents smarter. It's a protocol for tool connectivity. The agent still needs guardrails (max_iter, cost budget).