AppsFlyer lets marketers build their own AI agents and adds four MCP tools

What AppsFlyer announced

AppsFlyer, a company that measures how mobile app marketing performs, announced a broad expansion of its AI agent features, as Zawya reported. An AI agent is software that can carry out a multi-step task on its own, rather than just answer a question. The company pitches the release as a way for marketing teams to run agents with clear visibility into what those agents are doing.

The update has three parts. It widens a feature called Agent Hub so marketers can build their own agents. It adds the company's first agent that can make a change inside an advertising platform. And it adds four tools built on the Model Context Protocol, or MCP, an open standard that lets AI assistants connect to outside tools and data.

AppsFlyer's framing is that the question for marketers in 2026 is less about what agents can do and more about how much control teams have over them. The release is aimed at that concern.

Build-your-own agents in Agent Hub

In the expanded Agent Hub, a marketer describes a marketing performance workflow in everyday language, and the system builds an agent for it. That agent pulls data at the level of individual events, analyzes it, and sends findings to Slack or email on a schedule the team picks.

It can also dig into the cause of a change, such as a sudden jump in fraud, and suggest a next step. The company gives examples of the kind of work it has in mind: investigating fraud spikes, spotting creative fatigue (ads that people have stopped responding to), and recommending bid changes in Apple Ads.

The pitch is that marketers decide strategy based on what the agents surface, rather than handing decisions over.

An agent that acts, with a human sign-off

The most notable piece is the Apple Ads Share of Voice Agent. Share of voice is a rough measure of how visible your ads are for a given search keyword compared with competitors. This agent estimates it for each keyword and each country or region, and flags when it falls off target.

When that happens, it proposes a bid change and can push that change into Apple Ads straight from AppsFlyer. This makes it the first AppsFlyer agent that does more than advise. Nothing is applied until the marketer approves it.

It is not widely available yet. The agent is in closed beta, and interested customers go through their AppsFlyer customer success manager.

Four MCP tools for your own AI assistant

MCP is the part with the broadest implications. Instead of logging into a dashboard, a marketer can ask questions of their AppsFlyer data from an AI tool they already use. The four new capabilities are:

• Fraud data. Data from Protect360, AppsFlyer's fraud protection product, is now reachable. Examples include breaking down fraud by geography over 30 days, or finding which custom validation rules blocked the most traffic.
• Creative performance. Data is available per ad creative, covering spend, reach, conversion, retention and revenue cohorts. The aim is to find winning and tiring creatives faster and automate around them.
• Campaign link creation. The OneLink tool turns a plain-language campaign description into a configured link. Nothing is created until the marketer approves.
• Partner integration settings. Configuration, attribution windows, postback rules and event mappings can be read by any MCP client, so marketing and support staff can troubleshoot setup problems by asking questions.

AppsFlyer says all of these are available in its own AI assistant as well.

Who this affects and what to ask

The direct audience is marketing teams already using AppsFlyer, which describes itself as trusted by over 15,000 businesses. Sarah Maina, the company's regional manager for the Middle East and France, said the Gulf's gaming, fintech and e-commerce app companies are keen to use agents but want to understand the reasoning behind recommendations and approve actions first.

Inna Weiner, the company's VP of Product, put the shift this way:

"Marketing teams are moving from manual campaign orchestration to agentic execution."

This is a vendor announcement, and "trusted" is the vendor's word. The source describes approval steps for the two features that make changes, but says nothing about pricing, wider availability beyond the beta, or how actions are recorded. Those are sensible things to ask an account rep before connecting an agent to live ad spend.

The wider pattern

The announcement is a small data point in a bigger trend. MCP began as a way for developers to connect AI assistants to their tools. Now a marketing analytics company is using it to expose fraud, creative and setup data to whatever assistant a marketer prefers.

The approval-first pattern also matters. Reading data is low risk, while changing bids or creating links has real costs. Vendors are drawing that line explicitly, and buyers can reasonably expect to see it elsewhere.