A UAE startup's MCP server brings AI-assisted Facebook and Instagram campaign management to Claude, ChatGPT, and Cursor with mandatory human approval gates.
Markifact, a UAE-based AI marketing automation platform, released a Meta Ads MCP server on Friday that pipes Facebook and Instagram ad data directly into Claude, ChatGPT, Cursor, and any other client built on the Model Context Protocol. Every write action the server can execute requires explicit human approval before it affects a live account.
The launch targets a gap that has been visible since AI assistants started getting useful in enterprise workflows. Most third-party connectors for ad platforms are read-only: they surface performance data but cannot touch the account. Markifact's server crosses that line, enabling campaign drafts, budget adjustments, ad set creation, and creative builds through natural language, then holding all changes in a paused state until a person signs off.
According to Dispatch, supported operations span the full Meta Ads stack: reporting, multi-account analysis, image and video ad creation, carousel and catalog formats, custom audiences, lookalike audiences, and optimization preparation. Marketers can query which campaigns are burning budget without results, which creatives are showing fatigue, and which placements are outperforming, then prepare changes in the same session that go live only after human review.
What it actually does
MCP, the Model Context Protocol, is an open standard that lets AI clients call external tools and read structured data from third-party systems. Anthropic published the initial specification; both Claude and ChatGPT now support it natively, which is why a single MCP server surfaces inside multiple AI clients without requiring separate integrations for each one. The standard has gained significant traction in the past year as a de facto middleware layer between enterprise software and artificial intelligence assistants.
For advertising agencies managing multiple client accounts, the multi-account analysis capability is probably the most immediately useful feature. Querying performance across an entire portfolio from a single prompt, without switching dashboards, removes real operational friction. Supported ad formats cover the majority of what agencies actually run on Meta's platforms: image, video, carousel, and catalog.
The server does not appear to support autonomous bidding strategy changes or targeting updates independent of Meta's API constraints. Markifact notes optimization workflows are available where supported, which signals some limits tied to Meta's own access rules rather than any deliberate product decision.
Context: who buys this and why now
ChatGPT reached approximately 800 million monthly active users as of early 2026, according to Forbes. OpenAI is converting that reach into enterprise revenue aggressively: Yahoo Finance reported in January that the company appointed Barret Zoph to lead its enterprise push, specifically to close the gap with Anthropic in business accounts. CNBC noted in March that OpenAI plans to hire roughly 3,500 additional staff by year end, with a share devoted to helping businesses deploy its tools.
Third-party MCP servers exist because the AI labs are not going to build vertical-specific connectors for every ad platform, CRM, and project management tool. Markifact's product is a bet that marketing teams will manage campaigns inside their preferred AI client rather than log into a dedicated tool. Whoever ships functional connectors first in a category accumulates real switching costs, since workflows built on one MCP server are not easily ported.
The approval-gated model reflects where trust in artificial intelligence agents actually sits in 2026. Running an autonomous agent against live advertising accounts, where a misconfigured campaign can consume five figures in an afternoon, requires a reliability threshold no vendor has demonstrated at scale. Approval gates are the current compromise: AI does the reasoning, humans bear responsibility for the outcome.
That equilibrium may shift. As models improve at campaign strategy and fewer human approvals get overridden, pressure to remove the friction will build. The question is whether Markifact treats the human-in-the-loop as a transitional phase or as a permanent product differentiator.
FAQ
Q: What is MCP and why are marketing companies building on it?
A: The Model Context Protocol is an open standard that lets AI clients like Claude and ChatGPT call external tools without custom one-off integrations. It has become the dominant middleware for connecting enterprise software to AI assistants, and adoption accelerated sharply through 2025.
Q: Does the server make changes to live ad accounts automatically?
A: No. Every write action, including campaign drafts, budget changes, and new ad sets, is held in a paused state requiring explicit human approval before anything goes live.
Q: Which AI clients are compatible with the Meta Ads MCP?
A: Claude, ChatGPT, and Cursor are listed explicitly. Any client with native MCP support should work.
Q: How does this differ from Meta's own AI-powered ad features?
A: Meta's native AI tools are embedded inside Ads Manager. Markifact's server brings Meta Ads data and actions into whichever AI client the marketer already uses, without requiring them to work inside Meta's own interface.