X is letting advertisers pipe their campaign data straight into ChatGPT and Claude
&imwidth=600&imheight=450&format=webp&quality=medium "X is letting advertisers pipe their campaign data straight into ChatGPT and Claude")
X is letting advertisers pipe their campaign data straight into ChatGPT and Claude Photograph: (Wikimedia Commons)
X has launched a Model Context Protocol server for its advertising platform, letting marketers connect campaign data and analytics to outside AI systems including ChatGPT, Claude Code and Grok, and query performance in natural language. The same week, OpenAI moved to cut off a Musk-owned company — making the openness on one side and the exclusion on the other worth reading together.
X has opened its advertising data to rival AI systems. Advertisers can now connect campaign performance to ChatGPT, Claude Code, Grok or a custom agent, and ask questions in plain language.
What Was Built
The mechanism is a Model Context Protocol server for X Ads.
Model Context Protocol is a specification for connecting AI systems to external data and tools. Rather than each model maintaining a bespoke integration with each service, a service publishes one MCP server and any compliant system can use it. It was introduced by Anthropic and has since been adopted broadly, including by companies that compete with Anthropic directly.
For advertisers the practical effect is that campaign analytics stop being something you read in a dashboard and become something you interrogate. A marketer can ask why performance dropped in a segment, or what to change in a creative, inside whichever AI tool they already work in.
The Part Worth Noticing
Trending Stories
&im=FitAndFill=(700,400))
)
)
)
)
)
)
)
)
)
)
)
)
)
)
&im=FitAndFill=(700,400))
)
)
)
)
)
)
)
)
)
)
)
)
)
&im=FitAndFill=(700,400))
X is owned by Elon Musk, who also owns xAI, which makes Grok.
An MCP server is deliberately model-agnostic. By building one, X has made its advertising data equally available to OpenAI's models and Anthropic's as to its own — which is a commercially generous thing to do for competitors.
That is more interesting than it first appears, and it happened in the same week OpenAI told Cursor it would cut off model access on November 12 because SpaceX had bought Cursor's parent company. OpenAI's stated reason was that it could not be confident a Musk company would honour its terms of service.
So one Musk-owned platform is opening its data to OpenAI's models while OpenAI is closing its models to another Musk-owned platform. Both decisions are defensible on their own terms. Together they say something about how unevenly this industry is deciding who to interoperate with.
Why Open It At All
The commercial logic is not complicated, and it is worth stating because generosity is rarely the explanation.
Advertising platforms compete for budget, not for model loyalty. Anything that makes X's ad performance easier to analyse makes it easier to justify spending there, and a marketer who can interrogate X's numbers in the tool they already use is more likely to keep using X than one who has to open a separate dashboard.
The data being exposed is also the advertiser's own campaign data rather than X's proprietary user information. That is a meaningful distinction: X is not opening its audience graph, it is making a customer's own results portable into a customer's own tools.
Where MCP Has Got To
The larger story is the protocol rather than the product.
Eighteen months ago the question in AI was which model would win. The competitive ground has since moved to how models reach data and take actions, and MCP has become the connective tissue for that. Google has been pushing its own Agent Payments Protocol, the American standards body has been working on agent identity and permissions, and there is legislation before Congress addressing verifiable, task-bounded authorisation for agents.
Standards work is usually the least interesting thing happening in a technology sector and the most consequential. Whoever's protocol becomes default determines what agents can reach, what permissions look like, and who can audit what an agent did.
The Security Question Underneath
There is a caution that applies to every MCP integration and is worth attaching to this one.
Connecting an AI system to a live business account gives it read access to real data and, depending on scope, the ability to act. That is the same class of arrangement behind the finding that 65 per cent of organisations experienced a security incident caused by AI agents on their own networks in the past year.
An advertising integration is at the milder end of that. It is analytics rather than spend authority, at least as described. But the pattern of the past year is that permissions granted for analysis get widened to action once the analysis proves useful, and the security review rarely gets repeated when they do.
What To Watch
Whether the integration stays read-only, or eventually lets an agent adjust budgets and pause campaigns without a human approving each change.
That is the line between a better dashboard and an agent spending money. The first is a convenience. The second is the thing every enterprise security team spent this year discovering it was not ready for.
