Let an AI agent run your social media, safely

The problem with agents and social media

Hand an agent your social accounts and two things can go wrong: it posts something you would not have approved, and it posts so often or so fast that a platform flags the account. Both are hard to undo. The usual answer is to not let the agent touch publishing at all, which throws away most of the value.

How pendpost makes it safe

• A human approval gate. Every post is fail-closed. The agent creates drafts and schedules them; nothing publishes until it's approved. By default that's you, and auto-approve is owner-only and revocable.
• Anti-ban circuit breakers. A platform action block halts that lane and does not auto-resume, so an over-eager agent cannot get your account suspended.
• Caption brand-lint. Captions are checked against editable rules before they can publish, catching AI-writing tells and broken links.
• Local-first. It runs on your machine, your credentials stay in your .env, and nothing phones home.

What the loop looks like

You ask your MCP client (for example Claude) to plan a week of posts. The agent drafts and schedules them through pendpost. You review the queue in the dashboard or from the agent, approve what you like, and edit or reject the rest. pendpost publishes approved posts when due, using native scheduling where the platform supports it, and reads insights back. The agent does the busywork; the judgment stays yours.

Getting started

Run npx pendpost and try it in mock mode with zero credentials, then connect your MCP client as described on the MCP server page. Full setup is in the docs.