What MCP Is and Why It Matters for Engineers
MCP (Model Context Protocol) is an open standard that gives Claude Code structured access to external tools and data. Understanding its architecture — tools, resources, prompts, transport types, and context cost — is the foundation for using it effectively.
Quick Reference
- →MCP = open standard for connecting AI tools to external services (not Claude-specific)
- →Three primitives: Tools (callable), Resources (readable), Prompts (injectable)
- →Transport: stdio for local servers, HTTP for remote/cloud-hosted servers
- →Each enabled server injects all its tool definitions into the system prompt — context cost is real
- →ENABLE_TOOL_SEARCH: lazy-loads tool definitions only when Claude searches for them
- →Use /mcp inside a session to see server status, enable/disable, and reconnect
- →Disable servers irrelevant to the current task to reduce token overhead
What MCP Is
Model Context Protocol (MCP) is an open standard for giving LLMs structured, permissioned access to external tools and data sources. It was designed to be AI-client agnostic — the same MCP server works with Claude Code, Claude Desktop, Cursor, and any other MCP-compatible client. You build it once; it works everywhere.
From Claude's perspective, MCP tools look identical to built-in tools like Read, Edit, and Bash. Claude decides when to call them based on their name and description. There is no special syntax for invoking MCP tools — you describe what you want in natural language and Claude picks the right tool.
MCP is an open protocol maintained independently of Anthropic. A PostgreSQL MCP server you configure for Claude Code also works with other MCP-compatible AI tools. Your investment in MCP setup carries across the ecosystem.