OpenAI Adopts Rival Anthropic's MCP as a Standard

Technical consensus rarely comes from the competitor's camp. This week, OpenAI CEO Sam Altman announced on X that his company will adopt the Model Context Protocol (MCP), the open standard developed by Anthropic to connect AI assistants to the systems where data lives. OpenAI will build it into its products, including the ChatGPT desktop app.

The gesture carries weight precisely because of who's making it. OpenAI and Anthropic compete head-to-head in the large language model market. For OpenAI to adopt a protocol born at Anthropic amounts to an explicit acknowledgment that MCP has become a piece of infrastructure too useful to ignore.

What MCP is and what it does

The Model Context Protocol is an open source standard that helps AI models produce better, more relevant responses to certain queries. Put another way: a language model on its own only knows what it learned during training. MCP lets it fetch fresh information and act on external systems.

The protocol lets models draw data from sources like business tools and task-completion software, as well as content repositories and app development environments. Its key function is enabling two-way connections between those data sources and AI-powered applications, such as chatbots.

The design rests on two components. Developers can expose data through "MCP servers" and build "MCP clients" — apps and workflows, for instance — that connect to those servers on command. It's a plug-and-socket setup: one side offers the data access, the other draws on it.

Hence the analogy making the rounds in the industry: MCP as the "USB-C of AI." Just as that connector unified cables and ports that were once incompatible across manufacturers, MCP promises a single, standardized layer for linking any model to any data source or tool, without custom-coding integrations for every combination.

A standard that was already gaining ground

OpenAI's adoption isn't a bet on an experimental protocol — MCP had already built real traction. In the months since Anthropic open-sourced it, a range of companies have added MCP support to their platforms, including Block, Apollo, Replit, Codeium, and Sourcegraph.

That lineup isn't a coincidence. It's dominated by software development and developer-tooling names — precisely the territory where connecting a model to a company's code, documentation, and systems delivers immediate value. An assistant that can read your repository and your knowledge base simply performs better than one improvising from training memory alone.

Anthropic chief product officer Mike Krieger welcomed the move by his company's rival. "Excited to see the MCP love spread to OpenAI – welcome!" he wrote on X. "MCP has [become a] thriving open standard with thousands of integrations and growing. LLMs are most useful when connecting to the data you already have and software you already use."

That last line sums up the thesis behind the whole project: a model's value doesn't lie only in its size or reasoning ability, but in its access to the specific context of whoever is using it.

OpenAI's timeline

Altman was specific about the rollout. "People love MCP and we are excited to add support across our products," he said. "[It's] available today in the Agents SDK and support for [the] ChatGPT desktop app [and] Responses API [is] coming soon!"

The order of the rollout is telling. Support starts with the Agents SDK, the developer-facing toolkit for building agents — AI programs capable of chaining actions together to complete tasks on their own. That MCP is landing there first, ahead of consumer-facing interfaces, signals where the real stakes lie. OpenAI says it intends to share more about its MCP plans in the coming months.

Why it matters that two rivals are converging

In recent years, major AI companies have tended to build walled gardens: proprietary formats, exclusive integrations, ecosystems designed to lock users into a single provider. Two direct competitors adopting a shared standard points in the opposite direction.

For developers, the upside is concrete. If the same MCP server works interchangeably with Anthropic's models and OpenAI's, building an integration stops being a bet on a single vendor. The work gets done once and serves multiple models, which lowers the cost of switching between them and pushes companies to compete on model quality rather than customer lock-in.

This convergence arrives just as the industry pushes toward agentic systems — models that don't just converse but execute tasks by connecting to real tools and data. A useful agent, by definition, needs to touch external systems. Without a standard connection layer, every agent would have to reinvent its links to every service. MCP offers that common substrate, and its adoption by the market leader cements its status as the leading candidate to underpin the ecosystem.

There's an irony worth noting here. Setting a widely adopted open standard is itself a form of soft power: whoever designs it sets the rules of the field. Anthropic open-sourced MCP, and now its biggest rival is adopting it and helping extend its reach. The benefit is mutual — more integrations for everyone — but authorship of the standard still belongs to Anthropic, and that origin point carries long-term value in an ecosystem that's only just taking shape.