DATALYR

Ads MCP for Meta, Google and TikTok.

One MCP server lets Claude and Codex read which ads bring paying customers, and agents propose Meta, Google and TikTok ads in your own accounts.

Pick your platform.

One server, one URL.

Why DATALYR's.

  • It knows what sells.

    Verified revenue from your store or billing is credited to each ad. The agent reads it before it proposes anything.

  • You approve everything.

    Every proposal lands on one review page. A person selects Upload, and it is created paused.

  • The owner holds the money.

    Only the owner turns ads on, inside three daily spend limits. One Stop button pauses it all.

What it reads.

What the agent reads from DATALYR
It readsSo the agent knowsTool
Which ads bring paying customersAttributed conversions and revenue down to the campaign, ad set and ad.get_attribution
Verified revenueOrders, subscriptions and refunds from your store or billing, not the ad platform's estimate.get_revenue
Spend and platform resultsWhat each ad spent and what the platform reported for it, next to the real revenue.get_ads
The ads already runningAd accounts, campaigns, ad sets and ads, so a proposal fits what exists.list_ad_campaigns
What agents already didEvery ad action in an activity log the agent can read too.list_ad_activity

Built-in guardrails.

DATALYR guardrails for agent ads
GuardrailWhat it does
Created pausedEvery campaign, ad set and ad starts paused. Nothing spends until something turns it on.
One review pageThe agent never changes your ad account from chat. A person checks each proposal on the review page and selects Upload.
Access levelsEach connection is Read-only, Propose, Create paused or Publish live. Members are read-only.
Owner-only turn-onTurning ads on and raising budgets wait for the workspace owner, only in ad accounts where the owner switched live ads on. It is off by default.
Spend limitsThree daily limits set by the owner. Defaults: 1,000 per campaign or ad set, 5,000 of new spend per 24 hours, 20,000 running in total, in your reporting currency.
Only what agents madeArchive and delete touch only objects agents created through DATALYR.
One Stop buttonPauses everything agents turned on and cancels what waits on review pages. Ask the agent to stop and it does it right away.
Rejected-ads pauseThree rejected ads in 24 hours on one ad account pauses agents there until someone turns them back on.
Emails and activity logOwners and admins get an email whenever agent work starts spend, and every action is logged.

Questions, answered.

Contact us
  • An ads MCP server is a connection that lets an AI assistant such as Claude work with your ad accounts. DATALYR's lets the assistant read your attribution and revenue and propose Meta, Google Ads and TikTok campaigns, ad sets and ads, paused, to one review page where a person approves them.

  • Meta (Facebook and Instagram), Google Ads and TikTok. For each, the agent proposes campaigns, ad sets or ad groups and ads, and can turn on, pause and change budgets, with the owner approving anything that starts spend.

  • Claude and Codex read your analytics, attribution and revenue. Agent ads work with Claude, ChatGPT, Claude Code and Codex, and any MCP client, using one address: https://mcp.datalyr.com/mcp. Setup for Claude and Codex is in the DATALYR docs.

  • No. Everything is created paused and goes to one review page. Turning ads on or raising a budget is limited to the workspace owner, in ad accounts where live ads are switched on, within three daily spend limits the owner sets.

  • Revenue. DATALYR is an attribution platform: it credits verified revenue from Shopify, Stripe, Whop, RevenueCat and Superwall to the campaign and ad behind it. The agent reads that, so it scales the ads that bring paying customers, not the ones with the best platform-reported ROAS.

  • Plans start at $299 a month with a 7-day free trial. See the pricing page for current plans.

See what really drives revenue.

Real-time attribution, analytics and conversions for every channel, campaign and ad, ready for your team and your agents.

Get started