Skip to main content

Glossary · ai

MCP (Model Context Protocol)

Definition

MCP (Model Context Protocol) is Anthropic's open standard released in late 2024. It provides a common communication protocol between AI agents and external tools/data sources. Claude Code, Cursor, Zed, Continue support MCP — called 'the USB-C for AI'.

Published: 2026-05-05Updated: 2026-05-05

Detailed explanation

Problem: every AI agent (Claude, Cursor, Zed) had its own tool integration API — N agents × M tools = N×M integrations. MCP solution: shared protocol → each tool implements one MCP server, each agent runs one MCP client → N+M.

Components: MCP server (tool provider — DB, API, filesystem), MCP client (AI agent — Claude Desktop, Cursor), MCP host (orchestration).

Ready MCP servers (2026): filesystem, SQLite, PostgreSQL, GitHub, Slack, Google Drive, Notion, Brave Search, Puppeteer (browser automation), 30+ Anthropic-official servers.

Development: Python/TypeScript SDK ships custom MCP server in 1-2 hours. The server exposes resources, tools, prompts; the agent discovers and uses them.

Use cases

Custom AI agent + internal system integration

Multi-agent ecosystem (every agent uses tools)

Cross-tool workflow (Claude → DB → API → Slack)

Reduce vendor lock-in via open standard

Pros

  • +Open standard (Anthropic-led, community-driven)
  • +Tool ecosystem growing fast
  • +Reduces vendor lock-in
  • +Easy custom server (1-2 hours)

Cons

  • Still early (released late 2024)
  • No native OpenAI/Google support yet
  • Less mature than LangChain
  • Production-grade observability lacking

Related terms

AI AgentTool UseLLMAnthropic

Related services

Planning a project around MCP (Model Context Protocol)?

In a 30-minute discovery call we share a written architecture + cost + team recommendation tailored to your project.

Start a discovery call