Snowflake previews AI Gateway to let ad agents act on Meta and TikTok

Snowflake has introduced the AI Gateway for Advertising, now in preview, which links its Cortex Agents to Meta and TikTok through governed Model Context Protocol (MCP) integrations. According to Snowflake, agents can combine business data with live ad platform signals, recommend an optimization and carry it out under policy, with a campaign manager approving each step.
- Snowflake's AI Gateway for Advertising is in preview, with MCP integrations for Meta and TikTok.
- It is built on Snowflake's May 2026 acquisition of Natoma's enterprise MCP platform.
- Cortex Agents combine Snowflake business data with live ad platform signals and act under policy with campaign manager approval.
- PMG is an early customer, and more platforms are coming soon according to Snowflake.
- Each recommendation, its evidence and approval are stored in one shared record.
What the AI Gateway for Advertising does
Snowflake says AI in marketing is shifting from a tool that assists to an agent that acts. In advertising, it argues, the obstacle is not only fragmented workflows but fragmented context. An agent that sees only a platform's metrics may lack the business definitions or channel signals needed to know what matters, and can end up recommending something that misses a core KPI or does not suit the channel.
The AI Gateway for Advertising is meant to close that gap. It is built on Snowflake's May 2026 acquisition of Natoma's enterprise MCP platform and currently offers MCP integrations with Meta and TikTok. Snowflake says additional platforms are coming soon. PMG is named as an early customer.
Connected to Cortex Agents, the gateway brings together enterprise context, such as business definitions and historical performance, with live platform signals including delivery, catalog health and diagnostics.
How the agent loop works
Snowflake describes a full campaign loop. The agent reasons over cross-channel performance, diagnoses live status and settings, decides the best action, explains its rationale to a campaign manager and acts with permission.
The company says an agent can pair live signals such as delivery, catalog health and conversion diagnostics with data in the advertiser's own Snowflake environment, including transaction history, inventory and the business's definitions of return on ad spend (ROAS) and conversion. The ad platform shows what is happening in the channel, while the Snowflake data helps show whether it matters to the business. Both sit on a shared set of definitions and KPIs.
Two examples from Snowflake
In a retail scenario, a marketer sees ROAS fall on a live seasonal campaign even though spend is pacing correctly and click-through rate looks healthy. Snowflake says finding the cause would normally take a multiday investigation. With the gateway, the marketer asks why, and the agent combines ad platform performance with transactions, inventory and conversion-signal health. It identifies three factors:
- Purchase event quality declined after a checkout payload change
- Several promoted SKUs have warnings
- A high-spend product group has limited inventory
The agent then recommends a bounded action: reduce the affected prospecting ad set's budget by 20%, but not pause it, because the converting cohort has strong lifetime value (LTV).
In a media scenario, an email promoting a new series draws a strong response, with one audience segment driving unusually high viewing and subscription activity. The agent links email engagement, viewing and subscription data with live campaign performance. It finds that viewers who clicked the email and watched the first episode convert to subscriptions at a higher rate, and that this segment is underrepresented in the current paid audience. The recommendation is to increase ad frequency for that segment by 50% for the next seven days while keeping broader prospecting unchanged.
In both cases, Snowflake says the recommendation, evidence and approval are captured in one shared record, so the next reviewer can pick up where the investigation left off.
Customer and partner views, and what comes next
Mike Treon, Head of CTV Strategy at PMG, said that building semantic views in Snowflake Cortex AI made business context reusable, and that it now takes his team "minutes to get from question to insight" where it used to take days. He added that the AI Gateway means agents can now act on that insight inside PMG's Alli platform while a campaign is live.
David Kaufman, Global Head of Monetization Product Partnerships at TikTok, said the TikTok MCP Server lets developers and platforms build AI solutions that connect with TikTok.
For advertisers, the pitch is a way to move from diagnosing a metric change to acting on it without leaving a governed, auditable workflow. The product is still in preview, and the source gives no pricing or general availability date. Snowflake plans to show the gateway at Advertising Week New York on Wednesday, October 7, in a session titled "What AI Agents Need Before They Touch a Media Decision."
Snowflake: AI Gateway for Advertising: Governed AI Agents | Snowflake


