On August 24, X connected its ad account directly to language models. The new server runs on the MCP (Model Context Protocol) standard and gives third-party AI agents — Claude, ChatGPT, and other compatible tools — access to 23 functions inside X's ad account: reading campaign data, running analytics, launching and editing ads. Advertisers no longer need to open the X Ads interface at all; they can describe the task in plain text and let the agent assemble the campaign.
Going by the example X itself shared, this works quite literally: a user types something like "launch a campaign for the fall drop. $15k over two weeks, US only" and gets back a structured campaign with that budget, timeframe, and geography within minutes. X isn't first here — Pinterest, TikTok, and Snapchat each rolled out their own advertising MCP servers over the past year. But X is one of the largest platforms on that list, which is exactly why this move pushes the idea from niche experiment toward market standard.
For X itself, this is part of a broader bet: the company is rebuilding its ad business around video and AI tools, in line with Elon Musk's vision of X as a "super app" where advertising, payments, and messaging live inside one product. In that logic, MCP isn't a one-off feature — it's a way to lower the barrier for advertisers who previously didn't want to learn X's interface separately from Meta's or Google's. It's also a response to competitive pressure: Pinterest, TikTok, and Snapchat already offered agent-based ad access, and X didn't — which was starting to hurt its standing as a destination for ad budgets.
Before MCP, automating work inside an ad account meant one of two things: hiring a specialist who knew a specific platform's interface, or commissioning a custom integration against that platform's API, with upkeep every time the API changed. MCP removes that step — an agent configured once can work with any platform that supports the protocol, because they all speak the same command language. For an agency, that means not rebuilding an integration for X from scratch, but connecting an MCP client once and getting access to all 23 account tools at the same time.
A coffee shop with a seasonal menu: where the agent actually saves time
Take a chain of three coffee shops in Lisbon that runs a promotion for a new drink once a season. Previously, the marketer spent half a day building the campaign in X Ads by hand — picking geography by neighborhood, setting a two-week budget, preparing separate creatives for three age groups. With MCP, that same marketer gives the agent one sentence covering budget, timeframe, geography, and goal, and gets a draft back in minutes instead of hours. For a business this size, with no dedicated media buyer and every campaign assembled by hand in the evening, that's a real time saving, not an abstract efficiency claim. A sensible setup here is having the agent prepare the draft and show it to a person before launch — the marketer checks budget and geography in under a minute and hits "launch" themselves, rather than handing that button to the agent from day one.
Who's on the hook if the agent spends the budget differently than intended
MCP gives the agent write access, not just read access to reports — that's what separates it from a chatbot that only offers advice. The agent can change a campaign's budget and authorize spend without mandatory human sign-off at every step. If a third-party agent misreads a condition like "weekdays only," or confuses extending a deadline with doubling a budget, the money has already left the ad account by the time anyone notices — and the question of who's responsible, the platform, the agent's developer, or the advertiser who connected the tool without limits, stays unresolved. Neither X nor the other platforms with similar MCP servers settle this for the advertiser — their terms of service put that responsibility on whoever grants the agent access.
The practical takeaway for small and mid-sized businesses: start with the narrowest set of permissions possible — let the agent draft campaigns, but don't let it launch them automatically until the budget structure becomes predictable. It's also worth keeping a separate log of everything the agent does and setting a hard daily spend cap that can't change without a second approval. Permissions expand gradually, once it's clear where the agent gets things wrong and where it's faster and more accurate than a person.
This is exactly the approach we build for clients at Dayava — not "plug in AI and walk away," but automation with control points wherever a decision actually touches the budget.
If a similar solution for your business sounds useful, leave a request at dayava.pt/contactos/.
Source: MediaPost, "X Launches Advertiser MCP, Enables Third-Party AI"