Overview
- Model Context Protocol gives AI apps a shared way to reach tools and data, so teams do not need to build a new plugin for every model and service.
- Anthropic introduced MCP in November 2024, OpenAI adopted it in March 2025, and Google DeepMind followed, with tens of thousands of servers now available across the ecosystem.
- Developers build one MCP server per service and any compatible client can discover and call its tools, which removes the N×M tangle of custom connectors.
- Security researchers reported MCP risks that include remote code execution paths and supply‑chain style attacks, and Anthropic fixed flaws in its Git MCP server after separate reporting flagged them.
- Scans in 2025 found many servers with no authentication and broad permissions, so teams are urged to use strong auth, log and review tool calls, grant least privilege, and only build MCP servers for shared, repeat use.