Anthropic open-sources the Model Context Protocol (MCP)

What happened

Anthropic open-sourced the Model Context Protocol (MCP), which it describes as a new standard for connecting AI assistants to the systems where data lives: content repositories, business tools and development environments.

The technical details

From the announcement:

• the problem: models are “trapped behind information silos”, and every new data source needs its own custom implementation
• MCP is an open standard for secure, two-way connections between data sources and AI-powered tools
• the architecture: developers either expose data through MCP servers or build AI applications (MCP clients) that connect to them
• three components were released:
  1. the MCP specification and SDKs
  2. local MCP server support in the Claude Desktop apps
  3. an open-source repository of MCP servers, with pre-built servers for Google Drive, Slack, GitHub, Git, Postgres and Puppeteer
• early adopters named: Block and Apollo; development-tool companies Zed, Replit, Codeium and Sourcegraph were working with MCP

Why it matters

Before MCP, each agent framework defined tools its own way, so each integration had to be written once per framework. In my September 2026 study post I built a small MCP server and client with the official Python SDK. By 2026, an MCP-based benchmark (MCP Atlas) appears in Google’s Gemini 3.5 launch numbers, which shows how quickly it became shared infrastructure.