You manage social media for a handful of clients, and you've been wondering how far to trust an AI assistant near the accounts. A social post is public the moment it ships. One wrong one, sent by a tool that was only trying to help, and you're the person writing the apology. So when you hear that Agorapulse has launched an MCP, you read the announcement looking for the catch.
The launch was October 29. The Agorapulse MCP connects the platform to Claude, ChatGPT and Microsoft Copilot Studio. From inside those assistants you can pull reports, listening insights and inbox summaries, and draft posts from your own data.
The catch you were looking for is a feature that isn't there. Agorapulse, in the announcement's words, "chose not to give their AI assistants a 'publish now' action." Drafts land in Agorapulse for review. Every inbox reply is confirmed with you first. If an approval workflow covers a profile, the reply waits for a manager. You sign in with your own login, so the assistant "only sees the profiles and data that person can access," and the connector stores no data.
You notice the numbers in the announcement and you weigh them accordingly. "More than 30,000 users in 180 countries" is company boilerplate. The "about 80 minutes a month per client" comes from one agency coordinator quoted by the CEO. Those are vendor figures, and the post you're reading says to treat them that way. It also says the design decision holds up without either one: the cheapest safeguard against an agent posting the wrong thing is to remove the action that posts.
There's a wrinkle you'd want to know about. The announcement says scheduling, editing and deleting posts from the assistant are coming "in the coming days." That is when the approval model will matter most, because those actions reach further than a draft does.
So you check what the APIs.io catalog says, because the catalog is where you'd go to see what's actually published. The Agorapulse provider page lists 26 API pages. The workflows in the story map onto them: the Reports API for reporting, the Inbox items and Inbox reply APIs for triage, the Drafts and Calendar Notes APIs for drafting, and the Competitor API for benchmarks. The REST surface is already wider than the MCP. The Simple Scheduling API, the capability the MCP says is still to come, is already in the record. The agentic-access profile counts 22 operations, and 7 of them act on data.
Then comes the part where the catalog turns on its own scoring. Agorapulse has a Kin Score of 69.7, in the exemplar band, with access clarity at 100.0 and operational transparency at 76.3. But contract governance is 14.4, the lowest of any facet. Agent Readiness is 54.3. Delegated identity, protected resource metadata and dynamic client registration are all lit, which matches the per-user sign-in you just read about. The unlit dimensions are reversibility, idempotency and dry-run mode.
That gap is the honest finding. Agorapulse has built human approval into the product, but its published contracts don't yet say which actions wait for a person. For you, that difference is practical. The approval step protects you today because the company built it into the tool. It isn't something you can read in the contract and plan around.
What you take away is simple. You can let an assistant draft and summarize against your clients' accounts, knowing the action that would post publicly was left out on purpose. You treat the 30,000 users and the 80 minutes as the vendor's numbers. And when scheduling and deleting arrive in the MCP, you ask where the contract says which of those actions wait for a person, because right now that answer lives in the product and not on the page.
Read the original: https://apis.io/2026/11/13/agorapulse-mcp-ships-without-a-publish-now-button/