On February 2, Amazon Ads VP Paula Despins stood at the IAB Annual Leadership Meeting and announced that the Amazon Ads MCP Server is now in open beta. If you manage Amazon campaigns, that matters. If you manage Meta campaigns, the silence from Menlo Park matters just as much.
Model Context Protocol, the open standard developed by Anthropic that lets AI agents communicate with external software through a standardized layer, has become the connective tissue of agentic marketing. In just over a year since its release, more than 5,800 MCP servers exist for tools ranging from CRMs to analytics platforms to ad networks. The advertising industry took notice. Two of the largest platforms in digital advertising responded in completely opposite ways.
Amazon threw open its doors. Meta said nothing.
For marketing teams managing campaigns on both platforms, that divergence has real consequences right now.
Amazon Ads: the official open invitation
The Amazon Ads MCP Server connects AI agents directly to Amazon Ads API functionality. It acts as a translation layer that turns natural language prompts into structured API calls. Advertisers and partners can link Claude, ChatGPT, Gemini, Amazon Q, or any other MCP-compatible agent through a single integration and start managing campaigns from there.
That last part matters. Before MCP, connecting Claude to Amazon Ads required one custom integration. Connecting ChatGPT required a completely separate one. With a single MCP server, any MCP-compatible agent plugs in. Amazon built the infrastructure so the platform is accessible natively by any compatible tool, regardless of which LLM you're running.
Amazon is also not just exposing raw API access. The server includes pre-built workflows that orchestrate Amazon Ads capabilities into complete, multi-step operations. Creating accounts. Generating reports. Launching campaigns. Expanding to new locales. What once required a developer stitching together three or more API calls can now be initiated with a single natural language prompt.
What the Amazon Ads MCP Server enables
- Campaign management: Create, update, and delete campaigns via natural language, including end-to-end Sponsored Products setup in a single workflow.
- Cross-market expansion: Expand campaigns running in one country to new locales with a single prompt. No manual reconstruction.
- Performance reporting: Pull reports, query metrics, and surface insights without console navigation or custom SQL.
- Billing and account settings: Manage account-level configuration and access financial data through the same natural language interface.
- AI platform flexibility: Works with Claude, ChatGPT, Gemini, Amazon Q, and any other MCP-compatible agent. Bring your own LLM.
The business logic is straightforward. Friction in campaign setup is Amazon's enemy. Amazon competes with Google on search advertising and with every other retail media network on relevance. The more AI agents can plug into Amazon Ads and execute without friction, the more campaigns get created, managed, and optimized on Amazon's properties. First-party MCP infrastructure is not a developer convenience. It is a revenue strategy.
The efficiency numbers being reported in the field reflect that. Agencies adopting AI agents for Amazon campaign management are seeing 70–80% reductions in setup and management time. The MCP server automates the execution. Humans own the strategy. That is a division of labor most marketing teams can actually work with.
"The shift isn't toward AI-assisted advertising, where humans do the work and AI provides suggestions. It's toward AI-managed advertising, where agents execute workflows directly while humans focus on strategy and oversight."
There is also an infrastructure benefit worth noting. As Amazon's APIs evolve, AI agents remain compatible automatically. No code rewrites. No integration maintenance cycles. Amazon hosts and updates the server on behalf of advertisers. That is a meaningfully lower maintenance burden for any agency or in-house team running AI workflows at scale.
Meta: silence, third-party tools, and suspended accounts
Meta's position on MCP advertising automation is, officially, nothing. No announcement. No developer documentation. No MCP server of its own. There is no Meta equivalent of the IAB keynote moment.
What exists instead is a community-built ecosystem of third-party MCP servers for the Meta Marketing API. Open-source packages on PyPI and GitHub. Tools like pipeboard.co and GoMarble. These operate through Meta's existing Marketing API and, on paper, comply with its terms of service. Their maintainers are careful to note that these are unofficial third-party tools, not associated with or endorsed by Meta.
That qualifier has real consequences.
A notable thread has been circulating in the paid media practitioner community on LinkedIn. PPC consultant Malcolm Gibb, founder of Flowio and a veteran of iProspect and leading UK agencies, surfaced what he described as "quite a few stories" of Meta ad accounts being suspended after users connected third-party MCP tools. Practitioners Rahul Aroraa and Vaishnavi Mishra posted similar warnings. Mishra explicitly cautioned that connecting MCP to your Meta ad account may get your account banned. The posts attracted responses from other practitioners who had hit the same wall.
None of these suspensions came with a message from Meta that said "you used MCP." They never do. Meta's enforcement is automated and opaque.
Meta's Account Integrity policy explicitly flags creating or using accounts or entities through automated means, such as scripting, unless the scripting activity occurs through authorized routes and does not otherwise violate policies. The operative phrase is authorized routes. Third-party MCP servers, however well-intentioned, are not Meta-authorized routes.
Why Meta's systems likely flag MCP connections
- Unauthorized automation: Meta's terms restrict scripting or automated access that doesn't go through approved channels. Third-party MCP tools have not been designated as authorized routes by Meta, despite using the official Marketing API.
- Unusual API traffic patterns: MCP-powered agents can generate rapid, sequential API calls that look anomalous to Meta's risk detection systems. Those patterns look similar to what Meta associates with compromised accounts or unauthorized scraping.
- No official blessing: Amazon built its own server and actively encourages this use. Meta has made no statement endorsing MCP-based ad management. That ambiguity is not a gray zone. It is a gap in your coverage.
What makes this worse for advertisers is that Meta's enforcement doesn't distinguish between well-intentioned developers and bad actors. It flags patterns. Rapid budget increases, logging in from unfamiliar locations, unusual API traffic: all of it can trigger automatic restrictions. Even legitimate, read-only access can look like a compromised account to a system trained on fraud patterns.
The appeal process offers limited comfort. Numerous reports indicate that even when users provide valid justifications or evidence, the automated system often rejects the appeal. Usually leading to a permanent ban. For an agency managing client accounts, that is not a risk that belongs to you alone.
What these two platforms are actually betting on
Framing this as "Amazon good, Meta bad" misses the more interesting story. These are two companies making different bets about what role they want to play in the AI-driven advertising stack.
Amazon's approach is platform-first openness. By building an official MCP server and actively encouraging AI agents to interact with Amazon Ads, Amazon is saying: we want to be infrastructure. The platform choice, whether Claude, ChatGPT, or Gemini, doesn't matter. What matters is that campaigns get created, managed, and optimized on Amazon's properties. Low friction, open architecture, for a company that profits every time an ad runs.
Meta's posture reflects a different history. Meta has fought expensive battles over data access for years. Cambridge Analytica. The deprecation of third-party data. The IDFA fallout. The dismantling of broad targeting. The company's instinct, shaped over years of platform abuse, is to maintain tight control over how its data and systems are accessed. That instinct doesn't have to conflict with MCP. But it does mean Meta won't let a community-built server ecosystem operate without scrutiny, especially one with no official sanction.
There is also a structural asymmetry worth naming. Amazon's advertising moat is built on ease of execution at scale. Meta's moat has historically been its first-party audience data and targeting capabilities, not its developer experience. That shapes everything about how each platform responds to agentic automation.
| Dimension | Meta (Facebook Ads) | Amazon Ads |
|---|---|---|
| Official MCP Server | None. No official stance. | Yes. Open beta since Feb 2, 2026. |
| AI Agent Access | Third-party tools only. Unofficial. | Claude, ChatGPT, Gemini, Amazon Q. Officially supported. |
| Account Risk | Reported suspensions after MCP connections. | No risk. Officially sanctioned infrastructure. |
| Platform Philosophy | Controlled access. Closed ecosystem tendency. | Open infrastructure. Bring your own LLM. |
| Advertiser Clarity | Ambiguous. No guidance on MCP use. | Full documentation. Globally available to API partners. |
| Developer Ecosystem | Community-driven. Unofficial disclaimer required. | First-party, Amazon-hosted infrastructure. |
What to do right now
For agencies and marketing teams managing both platforms, the practical split is immediate.
On Amazon: The MCP server is real infrastructure you should be exploring. It automates execution workflows, not strategy. The decisions that drive Amazon advertising performance still require human expertise. The MCP Server makes executing those decisions faster. Use it for campaign setup, reporting, and market expansion. Keep humans on architecture, diagnosis, and cross-channel context.
On Meta: Hold. The community-built tools are technically impressive and their maintainers believe they operate within the Marketing API's terms. That may be true in a narrow legal sense. It doesn't prevent Meta's automated systems from flagging the traffic patterns and suspending the account. Until Meta issues official guidance or releases its own MCP server, the risk is real and the appeal process is opaque. Wait for clarity before connecting client accounts.
The broader question this raises for the advertising industry: can Meta afford this posture for long? Nearly half of marketing organizations are already using MCP connectors in AI assistants. Another quarter are incorporating MCP into their agents and automations. As AI-managed advertising becomes standard, platforms that make it easy to automate will attract advertiser attention and budget. Amazon has placed its bet publicly and in writing.
Meta's chapter remains blank. Whether that turns into an official server, a formal crackdown, or a policy clarification that changes nothing, we do not know yet. What we do know is that advertisers caught in the middle are making real decisions with incomplete information, and the cost of getting it wrong falls on them.
Sources: Amazon Ads official announcement (advertising.amazon.com, February 2, 2026) · AdWeek, AdExchanger, Digiday, and Marketing Dive coverage of IAB ALM 2026 · Meta Transparency Center, Account Integrity and Advertising Standards policies · LinkedIn posts from Malcolm Gibb, Rahul Aroraa, and Vaishnavi Mishra · Canopy Management MCP analysis · W Media Research · Averi.ai MCP overview. Community-reported Meta account suspensions are based on practitioner posts and represent anecdotal field reports, not verified causal relationships confirmed by Meta. Meta has not issued any official statement regarding MCP usage and ad accounts as of publication.