Key takeaways
In November 2024 Anthropic released an open standard called the Model Context Protocol. For a while it looked like one company’s plumbing for its own products.
Within eighteen months it had become the way the AI industry connects models to everything else. If your AI tools started connecting to each other over the past year, MCP is why. It’s the biggest thing in AI software since the models themselves got good.
What problem did it solve?
Before MCP, connecting a model to a tool was a custom job every time. Claude querying your database needed a Claude-specific connector. ChatGPT needed another, and Gemini a third. Each had its own login, data format and ways of breaking.
Engineers call this every-model-times-every-tool mess the N-by-M problem. MCP fixes it. Build one MCP server for each tool and one client for each AI host, and every combination works. It’s like USB-C replacing a drawer of chargers that don’t fit.
Build one MCP server per tool and one client per AI host, and every combination just works.
Why did it become the standard?
Lots of companies release open standards. Almost none become the standard. MCP did because of who adopted it, and in what order.
Within months OpenAI adopted it across its products, including the ChatGPT desktop app. Anthropic’s biggest rival backing it turned “Anthropic’s thing” into a neutral standard. Google DeepMind and Microsoft followed.
In December 2025 Anthropic gave MCP to the Agentic AI Foundation, a fund within the Linux Foundation co-founded with Block and OpenAI. That removed the single-vendor risk that makes companies wary of building on someone else’s standard.
By mid-2026 there were roughly 10,000 public MCP servers by the official registry’s count. Code-hosting sites had even more repositories tagged as servers.
The protocol reached in about 16 months a level of downloads that took React about three years. It’s an imperfect comparison, but it shows how steep the curve is.
| Category | Months to comparable adoption |
|---|---|
| MCP | 16 mo |
| React (early growth) | 36 mo |
Cite or embed this
Where did the hype get ahead of the facts?
The gap between those figures is the gap between “MCP has won” and “MCP is winning”. The same surveys named security as the main thing holding back wider use. That’s what you’d expect from a standard built to let models reach into your real files, databases and systems.
Building with AI in 2026? MCP is the connection layer. With around 10,000 servers published, most teams can connect AI to the tools they already use in an afternoon. Usually that means setting up an existing server, not writing one.
What turns a smart model into a useful agent is that connection layer. An agent can read your documents, query your systems and act. MCP is that layer becoming standard.
Frequently asked questions
What does MCP stand for?
Model Context Protocol. It's an open standard for how AI models connect to outside tools, databases and APIs through one shared protocol.
Who created MCP and when?
Anthropic released MCP as an open-source protocol in November 2024. In December 2025 it was donated to the Agentic AI Foundation, a fund within the Linux Foundation co-founded by Anthropic, Block and OpenAI.
Why is MCP compared to USB-C?
Before USB-C, every device needed its own cable. Before MCP, every link between an AI model and a tool needed its own custom integration. That's the 'NxM problem.' MCP turns it into one shared protocol, so any compatible AI host can connect to any compatible tool.
Which companies support MCP?
Anthropic (Claude Desktop, Claude Code), OpenAI (ChatGPT desktop), Google DeepMind, Microsoft, Salesforce, Block, Cloudflare and Replit, among others. Code editors like Cursor, Zed and Windsurf have MCP clients built in.
Is MCP actually used in production?
More and more, yes. But be careful with the numbers. A widely repeated claim of 78% enterprise production adoption wasn't well sourced. A cleaner 2026 survey put roughly 41% of software organisations in some form of production use. That's strong for a young protocol, but smaller than the hype.
Do I need to build an MCP server to use it?
Usually not. Roughly 10,000 public MCP servers already exist. Most teams connect their AI to tools they already use by setting up an existing server, not writing one. That's typically an afternoon's work, not a project.
Three-minute lesson on this Workflow or agent? Marketing course, advanced. Free, no sign-up.