Competing AI labs do not usually agree on infrastructure. They compete on models, on pricing, on who ships the next frontier capability first. So when Anthropic open-sourced a connectivity protocol in late 2024 and, within five months, OpenAI and Google had both adopted it for their own agent platforms, that was already unusual.
What happened a year later is the part worth understanding. Anthropic did not just let other companies use the Model Context Protocol. It handed control of the protocol to a neutral foundation, co-founded with two of its most direct competitors.
That is not how you would expect a company to treat something that gave it a head start.
Think of MCP like a USB-C port for AI applications.
Quoted directly from the MCP documentation's introduction page
The problem it actually solves
Before MCP, giving an AI application access to your files, your database, or a search tool meant a custom integration for that specific application. Build the same integration again for the next tool, and again for the one after that. MCP standardizes the connection instead: a data source, a tool, or a workflow exposes itself once, through an MCP server, and any MCP-compatible application can use it. It is the same connective layer an agent needs when you wire it into a steered agent loop, where the tools available to the loop matter as much as the model driving it.
It is a connection standard, not a specific product. That distinction is exactly why it was possible for competitors to adopt it without handing Anthropic any lasting advantage. Everyone gets the same USB-C port. What runs through it is still up to them.
Thirteen months, six moments
November 25, 2024
Anthropic open-sources MCP
The specification, SDKs, local server support in Claude Desktop, and a repository of MCP servers, released as an open standard from day one.
March 26, 2025
OpenAI adopts it
MCP support ships in the Agents SDK. Sam Altman says people love MCP and that OpenAI is excited to add support across its products.
April 9, 2025
Google follows
Google DeepMind CEO Demis Hassabis says Google will add MCP support to its Gemini models and SDK, exactly two weeks after OpenAI's announcement.
September 2025
The registry launches
The MCP Registry goes live and reaches close to 2,000 entries, a 407 percent increase from its initial batch that same month.
November 25, 2025
First anniversary
Thousands of active servers, up from a handful of experimental ones. More than 2,900 Discord contributors, with over 100 new ones joining every week, supported by 58 maintainers backing the 9-person steering group.
December 9, 2025
MCP joins the Agentic AI Foundation
Anthropic donates the protocol to a new Linux Foundation directed fund, co-founded with Block and OpenAI, backed by Google, Microsoft, AWS, Cloudflare, and Bloomberg.
What year one actually looked like
The MCP maintainers marked the first anniversary by publishing the numbers rather than a victory lap. The growth is the kind that is hard to fake with marketing.
2,900+
Discord contributors, with 100+ new ones joining every week
407%
growth in the registry versus its initial September 2025 batch
58
maintainers backing the 9-person steering group
The maintainers put it plainly in the anniversary post: it is hard to imagine that a little open-source experiment, a protocol to provide context to models, became the de-facto standard for that exact scenario in under twelve months.
Rivals, as co-founders
The clearest signal of intent came on December 9, 2025, when Anthropic donated MCP to the Agentic AI Foundation, a directed fund under the Linux Foundation. Anthropic's own stated reason: MCP has been committed to staying open source, community driven, and vendor neutral since its inception, and a foundation is how that commitment gets made durable rather than dependent on one company's continued goodwill.
Anthropic, OpenAI, and Block sit on the same table as co-founders. Google, Microsoft, and AWS, three companies that build competing AI platforms and cloud infrastructure, chose to support rather than fork. That is what an open standard succeeding actually looks like from the outside.
What this means if you are exposing your own system to agents
MCP is exactly the kind of standard worth betting on precisely because of who governs it now. A protocol run by one lab is one roadmap decision away from becoming a liability. A protocol co-governed by Anthropic, OpenAI, Google, and Microsoft is infrastructure, the same way TCP/IP or HTTP are infrastructure nobody worries will get pulled out from under them.
That matters for anyone deciding how an AI agent should reach their own product. A document generation API is exactly the kind of external system MCP was built to expose: a data source and a tool an agent needs access to, without a bespoke integration for every client that wants to call it. Building the connection once, on the standard the whole industry converged on, ages a lot better than building it six separate times. For the wider set of decisions this involves, our AI engineering guide covers how to think about tool access, context, and agent architecture together.
Frequently asked questions
Related Resources
Best Skills, Plugins and MCP Servers
A curated list of the MCP servers and tools worth actually installing in a coding agent.
Loop Engineering
Building with AI agents by running a steered goal loop, connected tools included.
Agents That Document the Product
Guidewright drives your live app through the Chrome DevTools MCP server to write real documentation.
TurboDocx API and SDK
The kind of external system a standard like MCP is built to give an agent access to.
Build the connection once, on the standard that stuck
A document generation API built for agents to call, whichever standard or model is driving them.