Can it act without you?
Watch forA write tool that executes the moment the model calls it.
SpritesEvery write arrives as a proposal. New campaigns start paused.
Four questions before an agent touches ad spend
None of them is how many integrations the connector has. Each is what a bad answer looks like beside the one Sprites gives.
Is there a record?
Watch forActions visible only in the platform’s change history, unattributed.
SpritesEach approved change is logged against the account, the person and the run.
Does revoking stop it?
Watch forA long-lived API key in a config file that keeps working after you leave.
SpritesRevoke the grant and the very next call fails closed.
One connector or one per platform?
Watch forA separate server, token and setup for every ad platform.
SpritesOne URL and one sign-in reach every channel connected in Sprites.
Agent Connectors for Marketing, answered
How is a connector different from an integration?
An integration is usually built once between two named products and ships in both. A connector is a general surface an agent discovers at runtime: it reads the list of tools the connector publishes and decides which to call. That is why one connector can serve many agents — Muse, Claude, ChatGPT, Cursor — without a separate build for each.
What separates a good agent connector from a bad one?
Three things, and none of them is the number of integrations on the box. Whether it can act without you or holds changes for approval. Whether there is a record of what it did, tied to the account and the person. And whether revoking access actually stops it, or whether a cached credential keeps working for a while afterwards.
Why does marketing need a different kind of connector?
Because the failure mode is spend rather than a bad row in a database. A connector that reads a CRM and gets something wrong wastes a moment; one that writes to an ad account and gets something wrong wastes budget, silently, until someone checks. That asymmetry is the argument for approvals on every write and for new campaigns starting paused.
Do I need a connector per platform?
Not if the connector sits in front of them. Sprites publishes one endpoint that reaches Google Ads, Meta, Microsoft Ads, LinkedIn, TikTok, Reddit, GA4, Search Console, Shopify, WordPress and Google Business Profile, because the platforms are connected once inside Sprites and the connector acts on those connections. One URL, one sign-in, every channel.
Which agents can use one?
Any that speak the Model Context Protocol. Thirteen clients are on the Sprites connector today with setup verified in each — Claude, ChatGPT, Gemini, Cursor, Codex, GitHub Copilot, Grok and more — and Meta’s Muse reaches the same endpoint through its Custom Connector flow.
What one connector reaches
Every channel below runs through the same endpoint and the same grant. Connect a platform once inside Sprites and the agent never touches a credential for it.
Google AdsGoogle Ads (official Google Partner OAuth)MetaMeta (Facebook and Instagram) AdsLinkedInLinkedIn Campaign ManagerTikTokTikTok Ads ManagerRedditReddit AdsGA4Google Analytics 4PostHogPostHog (personal API key and a selected project)SEOGoogle Search Console and your CMSAEOYour site, plus AI answer-engine trackingShopifyShopify AdminGBPGoogle Business ProfileWordPressWordPress (Application Password or the Sprites plugin)WebflowWebflow (a site API token with CMS, Pages and Assets access)FramerFramer (connected through the MCP plugin from the Framer Marketplace)Zoho CRMZoho CRM (Zoho OAuth, any Zoho data center)
Where this stands
The same plain statement the main guide makes, because it belongs on every page that tells someone to paste something.
Live today
- The Sprites connector is live at https://mcp.sprites.ai/mcp, over streamable HTTP with OAuth.
- Thirteen other AI clients are connected to it today, with setup we have verified in each.
- Every channel below is reachable through that one endpoint right now.
Not verified yet
- Muse builds custom connectors from a hosted URL, which is exactly the shape we publish — but we have not yet walked a Muse setup end to end ourselves, so the message above is not a verified guide.
- Early testers report that an OAuth sign-in takes more back-and-forth in Muse than a static API key does. Sprites signs in with OAuth, and we are testing that handoff now.
Common questions
Is an agent connector the same as an MCP server?
An MCP server is one way to build one, and currently the most portable: it is an open protocol, so a single server serves every client that speaks it. "Agent connector" is the category; MCP is the standard most of the category has settled on.
Can an agent connector work without API keys?
Yes, and it is better when it does. With Sprites you sign in once with OAuth and the connector acts on your workspace; there is no API key in a config file and no shared service account for several people to hide behind.
What should I check before connecting one to an ad account?
Whether it can apply a change on its own, what happens to a new campaign it creates, where the record of its actions lives, and what a revoke actually does. If any of those answers is vague, that is the answer.
Who builds agent connectors?
Mostly developer-tooling companies aimed at moving data between SaaS products. Very few are built for the case where the agent acts on money, which is the gap Sprites exists in.
Give your agent something worth asking about
Connect your ad accounts to Sprites once. Muse, Claude, ChatGPT or any other client that speaks the protocol reads them from there — and every change waits for you.