AdKit went live with a Model Context Protocol service that permits AI agents to execute campaign decisions directly inside Google Ads and Meta Ads Manager. The Singapore platform announced the integration without a staged rollout, giving marketers immediate access to programmatic budget allocation, creative rotation, and bid adjustments governed by large language models instead of human trafficking teams.

The service uses Anthropic's MCP standard to establish authenticated connections between conversational agents—Claude, ChatGPT, or custom enterprise models—and the API infrastructure of Alphabet and Meta. An agency strategist can now instruct an AI agent to launch a $50,000 Instagram carousel test across twelve markets, pull performance data at the creative-unit level, and reallocate spend toward the top three executions within a single natural-language session. AdKit handles OAuth, rate limits, and error handling; the agent sees a unified schema across both platforms. The company did not disclose pricing but confirmed the service is live for existing AdKit customers and available to new accounts under annual contract.

This matters because it removes the last manual layer between strategic intent and media execution. Holding companies have spent three years building AI co-pilots that summarize dashboards and suggest optimizations; AdKit skipped the suggestion layer entirely. The protocol architecture means an AI agent operating inside a family office's workflow can now treat paid social the way it treats calendar management—issuing instructions, verifying outcomes, adjusting without human approval loops. For luxury hospitality groups running seasonal acquisition plays across 15 to 25 markets, this compresses decision-to-deployment from days to minutes. The risk is equally compressed: a misconfigured agent could burn a quarterly budget in an afternoon, and neither Google nor Meta has published guidance on liability when an API key is held by a non-human actor.

The Model Context Protocol itself is six months old, released by Anthropic in late 2024 as an open standard for connecting AI systems to external data sources. AdKit is the first ad-tech platform to implement it for live campaign management, ahead of Trade Desk, Mediaocean, and the holding-company proprietary stacks. That timing suggests either aggressive product velocity or a calculated bet that MCP becomes the Rails-style protocol for agent-driven marketing infrastructure. Observers should note whether Google and Meta issue formal MCP adapters for their ad APIs in the next 90 days; if they do, AdKit's six-month head start compresses to zero.

Watch for three developments by mid-year. First, whether any of the six global holding companies announce MCP integrations with their own orchestration layers—Publicis and WPP both run large Singapore operations and have line of sight to AdKit's deployment. Second, whether luxury verticals with high creative variance and tight margin discipline—watches, spirits, automotive—begin requiring MCP compatibility in their agency RFPs. Third, whether the first public campaign failure attributed to agent error surfaces, and whether the liability falls on the marketer, the platform, or the model provider. AdKit has 12 enterprise customers according to its site; that sample size will produce an answer faster than anyone expects.

The protocol is live, the agents are credentialed, and the spend infrastructure is open. The question is no longer whether AI can manage campaigns, but whether the humans holding the budget have built the guardrails to survive the speed.