On 9 December 2025 Anthropic announced it was donating the Model Context Protocol, or MCP, to the Agentic AI Foundation, a directed fund under the Linux Foundation. Anthropic, Block and OpenAI co-founded the foundation, with Google, Microsoft, AWS, Cloudflare and Bloomberg among its supporters. The announcement also cited more than 10,000 active public MCP servers and adoption in products including ChatGPT, Cursor, Gemini, Microsoft Copilot and Visual Studio Code.
Why this is worth caring about
Every AI agent needs the same plumbing: a way to discover which tools exist, call them safely and receive structured results. Before a shared standard, each vendor invented its own, and every integration was bespoke. A common protocol does for agents roughly what standard interfaces did for earlier software: it lets tools and agents be built separately and still work together.
What it changes for what you build
- Build your tool layer once — expose your internal systems (ticketing, CRM, inventory) through a small, well-defined interface, and let different agents and models use it. That separates your durable asset from a model choice you'll revisit.
- Reduce lock-in — if the tool interface is standard, changing the model or the agent framework doesn't mean rewriting every integration.
- Govern at one point — permissions, logging and rate limits can be enforced at the tool boundary, which is where an auditor will look anyway.
What it does not solve
A standard connection is not a safe connection. Every tool an agent can call is an action it can take, and every third-party server you connect is code you are trusting. Treat each one like a new integration with a new vendor.
- Least privilege — give each tool the narrowest access that works, scoped per user wherever possible.
- Vet what you connect — prefer servers you control or can audit; review anything third-party before it touches production data.
- Approval gates — decide which actions need a human before they run, and enforce that in the tool layer instead of the prompt.
- Logging — record every call with its inputs and outputs, so you can explain what the agent did.
The practical takeaway
If you are planning agent work, ask early which of your systems should be reachable through a standard tool interface, and who owns the permissions for each. If you are buying, ask whether the product speaks the standard or locks you into its own connectors. Standards are unexciting by design, and that is their value: they make the exciting parts replaceable.