Agents, API and MCP
Letting an AI agent post for you, and the three things that stop it embarrassing you.
How do I let an AI agent post for me?
Create an API key with the agent role under Project settings → Agent — that tab also has ready-to-paste connect snippets for Claude, Cursor and plain REST. Both the hosted MCP connector and the REST API are included in every plan.
The developer docs have the whole thing — a quickstart, all the MCP tools, and the REST reference.
What stops an agent posting something embarrassing?
Three things, and they're the reason xpost exists. Guardrails run over every draft an agent writes, exactly as they do for yours. In copilot mode the agent's post lands in Approvals and waits for you. And an agent key can never approve anything, including its own work.
An agent can also check a draft against your guardrails before creating a post, so a well-behaved one fixes its own wording rather than filling your queue with rejects.
Still stuck? Ask a person.
Email support@xpost.to. If it's about a post that went wrong, the fastest thing you can send is the link to it from Receipts — that page already has the platform's own error text on it.
Building on the API or wiring up an agent? The developer docs go deeper than these pages do.