I keep checking this list because it keeps changing. When I last went through every scheduler's documentation in August, Publer and Typefully had no MCP server. Both shipped one since. Anything you read on this subject that does not carry a date is probably wrong by now.
So here is the current state, checked on 1 October 2026 against each vendor's own documentation rather than against a roundup. Where I could not confirm something from the vendor themselves, the table says so instead of guessing.
I build one of the tools on this list. That section is marked, and the rest is a straight answer to the question.
The list
Ten of the sixteen schedulers I track now ship a first-party MCP server. Eight months ago it was four.
| Scheduler | MCP server | Notes |
|---|---|---|
| Buffer | Yes | Free on every plan, including the free tier |
| Metricool | Yes | Works on a free account |
| Planable | Yes | |
| Postiz | Yes | Open source, works self-hosted too |
| Hootsuite | Yes | Three separate connectors |
| Blotato | Yes | Hosted |
| Mixpost | Yes | Pro and Enterprise licences, self-hosted |
| Publer | Yes | Hosted, Business and Enterprise plans |
| Typefully | Yes | Hosted at mcp.typefully.com, OAuth |
| RecurPost | Yes | Local server, 9 tools |
| planpo.st | Yes | Hosted or npm, 47 tools, includes revenue |
| Sprout Social | Unconfirmed | Reported on Advanced, not stated by the vendor |
| Post Bridge | Unconfirmed | Agent posting advertised, protocol not named |
| Later | No | Nothing in their own documentation |
| Loomly | No | Nothing in their own documentation |
| SocialBee | No | No public API either, which is the deeper reason |
| Post Planner | No | Nothing in their own documentation |
A date on this kind of list is not a formality
Two of the eleven yes answers above were no answers eight weeks ago. If you are reading a comparison that does not say when it was checked, assume it is describing a market that no longer exists. Mine says 1 October 2026, and it will be wrong eventually too.
First-party server or a bridge
This distinction causes most of the confusion, because a search for "socialbee mcp" returns plenty of results suggesting it exists.
A first-party server is built and run by the scheduler. It knows the product's own concepts, and the tools map onto things the product actually does.
A bridge, usually Zapier MCP or Composio, wraps whatever public API the tool exposes and presents it as MCP. It genuinely works, and it is thinner: you get the actions the bridge chose to expose, an extra subscription, and an extra place for things to break.
SocialBee is the clearest case. You can reach it through Zapier MCP, so "SocialBee MCP" is a real phrase. But SocialBee has no public API at all, by their own documentation, which limits what any bridge can do. Loomly is the same shape.
If a vendor's own docs do not mention MCP, you are looking at a bridge, whatever the search result says.
The question that matters more than yes or no
Once you have narrowed to tools that do have a server, almost everyone compares tool counts. That is the wrong number.
The useful question is what the agent can read back. Nearly every server on this list can write: draft a post, schedule it, publish it. Far fewer can tell the agent what happened afterwards, and almost none can connect a post to money.
The agent drafts and schedules. You find out how it went by opening the dashboard yourself, which puts a human back in the loop you were trying to close.
The agent can pull reach, clicks and engagement, so it can answer 'what worked' without you building a report.
The agent can answer 'what earned', which is a different question and the one most people actually care about.
That third column is where the list gets short. RecurPost's nine tools cover posting, libraries, history and AI drafts, and none of them return payment data, because RecurPost does not track it. The same is true of every other server above except ours, and that is a statement about what the products measure rather than about the protocol.
Notable entries
Buffer is the easiest recommendation if you want to try MCP without committing anything. The server is free on every plan including the free tier, so you can connect it this afternoon and see whether agent-driven posting suits how you work.
Postiz and Mixpost both expose MCP on their self-hosted versions, which is the only way to get agent access without a subscription to anyone. You pay in server maintenance and in registering your own developer application with each social network, which takes longer than people expect.
Metricool works on a free account too, and leans analytics, so the agent can read performance back.
Publer and Typefully are the two that changed recently. Publer puts it on Business and Enterprise, set up under Settings, AI and Automations, MCP. Typefully's is hosted with OAuth, which is nicer than pasting an API key and worth noting because very few are doing it that way yet.
SocialBee, Loomly, Later and Post Planner have nothing first-party. If you are committed to one of them and want agent access, a Zapier MCP bridge is your route, and you should expect it to feel like one.
How to actually connect one
The mechanics are the same whichever you pick, and shorter than people expect. You add a block of JSON to your client's config, or run one command, and the tools appear in the next conversation.
The fiddliest part is picking the right transport key, which is silently different for local and remote servers and generates no error when you get it wrong. I wrote that up in detail for the client people seem to struggle with most, in how to add an MCP server to Gemini CLI. The same shape applies to Claude Code, Claude Desktop, Cursor and Codex, and the per-client configs are in the agent docs.
If you want the wider version of this, covering what to ask for once the tools are connected and the guardrail to set before the first write, that is in how to build a social media agent.
Where I fit in this
My disclosure, so you can discount it accordingly: planpo.st is mine.
Our server exposes 47 tools across seven networks, hosted or as an npm package, and the part I would actually argue for is the revenue tools. The agent can ask what a post earned, because Stripe, Shopify and RevenueCat connect to the same account, so the question "did any of this sell anything" has an answer in the same conversation where the posts were scheduled.
What it does not do, plainly: no self-hosting, fewer networks than Postiz, no free tier at all, and no unified social inbox. Buffer's free MCP server is a better first experiment than paying me $19.99 to find out whether you like working this way. If you then want the revenue half, here is the comparison.
When none of this matters
MCP is a distribution channel for your attention, and it only pays off if you were going to delegate the work anyway.
If you write every post yourself because the writing is the point, an MCP server changes nothing about your week. If you already batch a week of posts in one sitting and the bottleneck is deciding what to say, an agent does not fix that either. Where it earns its place is the repetitive half: the platform variants, the scheduling arithmetic, the monthly "what happened" report that never gets written.
Pick the tool for the scheduling first. The MCP server is a tiebreaker, not a reason.