AdKit launches MCP service so AI agents can directly run Google and Meta ad campaigns
- AdKit said on April 30 it launched an MCP service that lets AI agents draft and manage Google Ads and Meta Ads campaigns. - The sharpest detail is the workflow: every agent action stays in draft until approval, and AdKit says 500-plus marketers already use it. - That matters because ad tools are becoming agent-ready fast, but advertisers still need guardrails before giving bots write access.
Advertising software is turning into agent software. That is the real story here. AdKit, a Singapore-based ad platform, said on April 30 that it launched an MCP service connecting AI agents directly to Google Ads and Meta Ads accounts — so a marketer can describe a campaign in plain English and have an agent build the draft. But the important part is not just the connection. It is the control layer sitting in between. (financialcontent.com) ### What did AdKit actually launch? AdKit launched a Model Context Protocol service for Google Ads and Meta Ads. MCP is the standard a lot of AI tools now use to let models call outside software and data sources. In practice, that means tools like C(financialcontent.com) reporting queries. (financialcontent.com) ### Why is that different from a normal ad API? A raw API connection gives software direct access. That is powerful, but also risky. Ad accounts are full of accumulated signal — pixel history, search data, bidding models, audience settings. One bad (financialcontent.com)irst in a review dashboard, with permissions set at a granular level, and nothing goes live until a human signs off. Basically, it is trying to make “AI can operate the account” feel less like handing your credit card to a sleepwalking intern. (financialcontent.com) ### What jobs is the agent supposed to do? The target is the repetitive operating work that eats a media buyer’s day — campaign setup, ad set creation, budget adjustments, targeting changes, pulling performance cuts, and assembling recommendations. (financialcontent.com)s not the same as saying the agent replaces strategy. It is more like compressing the mechanical part of the workflow. (financialcontent.com) ### Are there any numbers behind it? A few. AdKit says the service is already being used by more than 500 marketers and agencies. Early users, the company says, have cut campaign setup from roughly 45 minutes of manual work to a single prompt. It al(financialcontent.com) company-supplied numbers, so treat them as directional — but they tell you what AdKit thinks the wedge is: speed, lower agent cost, and safer execution. (financialcontent.com) ### Why launch this now? Because the plumbing is finally showing up. Google announced official MCP support for Google services in December 2025, with managed remote MCP servers instead of fragile local setups. At the same time, advertisers have been(financialcontent.com)ng the model, not owning Google or Meta, but owning the handoff between agent and ad account. (cloud.google.com) ### What is the catch? The catch is that ad buying is not just button-pushing. Good performance marketing depends on judgment — offer quality, creative taste, audience nuance, measurement discipline, and knowing when not to trust a recommendation. A draft-first workflow helps with operational safety, but it does not solve strategic quality by itself. If the prompt is bad, the draft can still be bad. (financialcontent.com) ### Where does this go next? AdKit says it plans to expand to TikTok, LinkedIn, Reddit, and X. If that happens, the bigger idea becomes clearer: one agent layer sitting across the fragmented ad stack, with humans reviewing the output instead of doin(financialcontent.com)rkflow compression that advertisers will actually trust. (financialcontent.com)
