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'.
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
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