Built-in agent · no setup

An AI social media manager that reports to you

Tell it what happened. It writes the post for every network, makes the image, checks the limits, picks the slots — and then waits for you to say yes.

Try the agent freeHow agent sessions work
Agent session
We shipped multi-workspace support. Get it out today and tomorrow morning.
Reading your brand kit and connected profiles…
list_brandings · get_connected_accounts
Four drafts ready — LinkedIn (long), X (280), Threads, Bluesky. Pre-flight is clean; 8 credits total. Publish LinkedIn + X now, queue Threads + Bluesky for 09:15 tomorrow?
preflight_post
Yes, but make the X one less corporate.
Rewritten. Awaiting your approval to publish.
create_post · awaiting approval
The job description

Six things it takes off your desk

Turns one idea into seven posts

You give it the news — a launch, a hire, a blog link. It writes a 280-character version for X, a longer narrative for LinkedIn, a caption for Instagram, and trims to 300 graphemes for Bluesky. Same message, native voice on each network.

Makes the image too

The agent can generate a post image from a prompt and attach it automatically. Point it at a brand kit and the visual style follows your reference image rather than drifting somewhere generic.

Plans the week, not just the post

Ask for five posts across the next seven days and it lays them out on the calendar with sensible spacing, in the timezone your audience actually lives in.

Checks before it commits

Every draft goes through pre-flight: character limits, disconnected profiles, missing media on networks that require it, and the credit cost. You see the verdict before anything is queued.

Cleans up its own failures

When a delivery fails, the agent can read the upstream error, fix the cause — a shortened caption, a replaced image — and retry just that profile.

Never posts behind your back

Anything that publishes surfaces as an approval first. Approve it, edit it, or reject it. Read-only work — reading the queue, checking account health — runs without interrupting you.

Autonomy with a brake

The difference between an assistant and a liability

An agent with unrestricted publishing rights on your company LinkedIn is a bad idea, however good the model is. PostMCP splits what the agent does into two classes and treats them differently.

Roles and approvals for teams
Runs freely
  • ✓Read the post queue and its statuses
  • ✓List connected profiles and page IDs
  • ✓Check which tokens are about to expire
  • ✓Read brand kits and workspaces
  • ✓Dry-run a post against limits
Needs your approval
  • !Publish a post immediately
  • !Schedule anything into the queue
  • !Edit copy or targets on a queued post
  • !Move a post to a different slot
  • !Delete a scheduled post
Teach it once

A brand kit is what stops the output sounding like everyone else’s

Tone of voice, target audience, the words you always use and the ones you never use, plus a style image for generated visuals. The agent reads the kit before it writes a single line.

Brand voice AI

Questions people ask before handing over the keys

What does an AI social media manager actually do?
It handles the work between having an idea and having it live: writing per-network copy, generating or picking an image, checking each network’s limits, choosing slots on the calendar, publishing, and retrying anything that failed. PostMCP’s agent does all of that against your real connected accounts, not a simulation.
Will it post without asking me?
No. Every action that publishes, edits or deletes surfaces as an approval you accept or reject. Read-only actions — listing the queue, checking which tokens are expiring — run freely, because they cannot change anything.
Does it sound like me or like a chatbot?
You teach it once. A brand kit holds your tone of voice, audience, guidelines, recurring keywords and a style image. The agent reads that kit before writing, so the output carries your phrasing rather than generic marketing filler.
Can I use my own AI instead of the built-in agent?
Yes. PostMCP is also a Model Context Protocol server, so Claude, ChatGPT or Cursor can drive the exact same tools. The built-in agent is the zero-setup path; MCP is the bring-your-own-model path. Neither is required to use the scheduler manually.
What does it cost to run?
Two meters. AI tokens cover the agent thinking and writing; publishing credits cover deliveries — 1 credit per profile, 5 for X, plus a 50-credit surcharge on posts containing a link. The Free plan includes 100k tokens and 20 credits a month so you can see the shape of your usage before paying.
Can a team share one agent?
Yes. The agent works inside a workspace, and workspace members with the Scheduler role or above can use it. Everything it does lands on the shared calendar and in the shared activity log.

Give it one sentence and see what comes back

Start free — 100k AI tokens