The short answer
MCP — the Model Context Protocol — lets an AI client talk to a service directly instead of you copying context back and forth. We are not the first scheduler to ship one: Metricool has one on its free tier, Planoly lists one from $14/month, and Vista Social bundles one at $149/month. So the interesting question is not whether a scheduler has MCP. It is what the tools let an agent do, and what happens when it gets something wrong.
Connecting it
One line, then a browser login. No API key to generate, paste or leak:
Authentication is OAuth 2.1 with PKCE and dynamic client registration, bridged to the same login you already use in the browser. API keys still exist for headless and CI use, but they are the fallback rather than the front door.
What the 67 tools actually cover
Roughly six groups. The one that makes this different from an analytics bridge is the first — these are the same editing operations the browser editor uses, exposed over the protocol.
- Edit the timeline — overlay add/update/delete, split, cut, silence removal, frame render.
- Captions — generate, restyle, export.
- B-roll and media — stock search, insert, transcript search, uploads.
- Publish and schedule — propose slots, draft posts, trigger a publish (confirm-gated, see below).
- Growth and strategy — content streams, issues, agents, memory, trends.
- Account — credits, budget, brand, projects, capability discovery.
The part that matters: it cannot post as you
"An AI that can post to your social accounts" is a thing people are right to distrust. So publishing is preview-then-confirm at the protocol level, not in the UI.
The first call to schedule or trigger a post refuses to act. It returns a draft and a preview link with a stable `CONFIRMATION_REQUIRED` response, and only publishes when called again with an explicit confirm flag. That applies to every caller, including our own code — there is no internal path that skips it.
The difference between "an agent can post as you" and "an agent can propose a post you approve" is one boolean, and it is the whole reason this is safe to turn on.
- Scoped — a token sees one owner and one active workspace, never another.
- Revocable — keys revoke instantly; OAuth tokens are per-client.
- Metered — AI-heavy tools spend credits, and the balance is readable over the same protocol.
Is this actually useful, or a demo?
Honestly, it is early. The workflow that has earned its keep so far is a boring one: point the agent at a long recording, have it search the transcript for the segments worth clipping, and get back proposed posts with times. Every one is reviewed before it goes anywhere. That is a genuine hour saved per long video.
What it is not is autonomous, by design. The agent proposes and a person decides — we would not ship it any other way while it can spend money and publish in your name.
Questions people actually ask
Which AI clients work with this?
Anything that speaks MCP over HTTP — Claude Code and Claude Desktop are what we test against. The connect command is one line and authentication happens in the browser.
Can the AI publish without asking me?
No. Scheduling and publishing tools refuse on the first call and return a draft plus a preview link. They only act when called again with an explicit confirmation flag, and that rule applies to every caller including our own services.
Do I need a paid plan?
No — MCP works on the free plan. AI-heavy tools consume credits, and new accounts get 50 starting credits. There is no recurring monthly credit allowance on the free tier, so sustained AI use eventually needs a paid plan.
Are you the only scheduler with an MCP server?
No, and anyone claiming that is not checking. Metricool, Planoly and Vista Social all ship one. What differs here is that ours works on the free plan, includes timeline-editing tools rather than analytics alone, and gates publishing behind explicit confirmation.