How AI agents work

In plain language: what you're connecting, what it can do, and why it's safe to let it.

What it is

Apps like Claude Desktop and Cursor can plug into outside tools using a standard called MCP (Model Context Protocol). PostStream publishes an MCP connector — a small bridge that lets the AI act on your PostStream account when you ask it to, and only within the permissions you grant.

You don't need to understand the protocol. You either paste one config block into your AI app, or use the one-click “Connect” button — both are covered in Connect an agent.

What it can do

Once connected, you can ask the agent, in normal sentences, to:

List your connected social accountslist_accounts
Write a post and schedule it for a date/time (or save a draft)schedule_post
Reschedule, re-target or edit a draft/scheduled postupdate_post
Cancel a scheduled post or delete a draftcancel_post
Retry a post that failed on one or more accountsretry_post
See what's in your scheduled queuelist_queue
Check a post's per-account statuscheck_post_status
Explain why a post failed (the real reason)explain_failure
Get the link to connect a new social accountconnect_account

You choose which of these it can do when you create the token — read-only, or read and write.

How it works

You

Ask your AI app to do something with your posts.

The connector

Translates that into an authenticated call to PostStream — it stores nothing.

PostStream

Runs the exact same service code the dashboard uses: checks your token, your subscription, your permissions, then acts.

Because it's the same code path as the dashboard, a post scheduled by the agent is indistinguishable from one you scheduled by hand — and it's covered by the same “did it actually publish?” confirmation.

Safety

It never sees your social passwords. Those live with the platforms; the agent only ever talks to PostStream.
Nothing publishes without your say-so. The agent drafts and schedules; you approve.
You can revoke it any time. One click in Settings → AI agents and the token or connection stops working on its next request.
Permissions are per-token. Give an agent read-only access if that's all it needs.

You need an active subscription

Creating a token or connecting an agent requires a paid PostStream plan. Beta testers can be added to a whitelist by an admin instead — ask your PostStream contact.