Model Context Protocol (MCP): Anthropic releases the open standard for agents

2024-11-25 · Publisher : Anthropic · Domain : Artificial Intelligence

Anthropic releases the Model Context Protocol, an open standard for connecting AI assistants to tools and data. Within a year, MCP becomes the reference protocol for agents, adopted by OpenAI, Google and Microsoft.

What changes

  • A single client-server protocol replaces custom integrations between each model and each data source.
  • Shipped with a specification, SDKs, support in Claude Desktop and ready-made servers (Google Drive, Slack, GitHub, Git, Postgres, Puppeteer).
  • Early adopters: Block and Apollo, plus the developer tools Zed, Replit, Codeium and Sourcegraph.
  • OpenAI adopts it in March 2025, followed by Microsoft and Google: MCP becomes a de facto standard.
  • In December 2025, Anthropic hands it to the Agentic AI Foundation, under the Linux Foundation.

On November 25, 2024, Anthropic open-sourced the Model Context Protocol (MCP), a protocol that standardizes how an AI assistant accesses tools, files and data. Anthropic described it as “a universal, open standard for connecting AI systems with data sources.” ### What happened - Anthropic released the specification and software development kits (SDKs), first in Python and TypeScript. - The Claude Desktop app supported local MCP servers. - An open source repository offered ready-made servers for Google Drive, Slack, GitHub, Git, Postgres and Puppeteer. - Block and Apollo integrated it, and developer tool makers Zed, Replit, Codeium and Sourcegraph were working on it from day one. ### How it works | Role | Function | |---|---| | Host / client | The AI application (assistant, IDE, agent) that connects to servers | | MCP server | Exposes tools, resources and prompt templates | | Protocol | JSON-RPC messages, local or remote | The same MCP server works with any compatible client: the “N models × M tools” problem is reduced to one integration per tool. ### Why it is a turning point MCP arrived just as models were becoming reliable enough to act. Within months, it became the common language of agents: OpenAI adopted it in March 2025, in its Agents SDK and then its products, and Microsoft and Google followed. In December 2025, Anthropic handed the protocol to the Agentic AI Foundation, a Linux Foundation fund, for neutral governance. ### Caveats Opening a model to tools widens its attack surface: prompt injection through third-party content, malicious servers, overly broad permissions. Early versions had no standardized authorization for remote servers; OAuth was added in the 2025 revisions. ### Key takeaways MCP did for agents what REST APIs did for the web: a shared contract. For a company, exposing its tools and data over MCP makes them usable by any compatible assistant, provided security is under control.

Links

Tags : IA, Agents, MCP, Open Source, Anthropic, Interopérabilité