MCP Social Media Scheduler — Calendar for You, Tools for Your AI

To schedule social media posts with MCP, you connect a scheduling tool’s MCP server to an AI client (Claude, ChatGPT, Cursor, Hermes, OpenClaw, Claude Code, or anything else that speaks the protocol), then ask the AI in plain language to schedule the post.

The AI calls the server’s create_post tool, the post lands in your scheduler’s calendar, and it publishes at the time you asked for. Setup takes about five minutes.

The rest of this article shows exactly how, using SchedPilot’s MCP server, and covers what to do once it’s working.

If you already know what MCP is, skip to the setup. If not, one paragraph.

MCP (Model Context Protocol) is a standard that lets an AI model call tools on an external service. Without it, Claude can write you a LinkedIn post but can’t do anything with it. With an MCP server connected, Claude gets a small menu of actions the service exposes, like “list connected accounts” or “schedule a post,” and it picks the right one based on what you asked. There’s a fuller explanation in our piece on what a social media MCP server is, but you don’t need the theory to use it.

What you need before you start

An MCP client. Claude Desktop and Claude Code are the most common. Cursor, Windsurf and ChatGPT (via custom connectors on paid plans) also work. The same server plugs into all of them.

A scheduling tool with an MCP server. As of late 2026 that list is short. SchedPilot has one, SocialPilot launched one recently, and a handful of API-first tools like Ayrshare can be wrapped in a community server. Buffer, Hootsuite and Later don’t offer MCP yet, which is odd given how loudly they talk about AI. We ranked the options in our best MCP servers for social media roundup; this guide uses SchedPilot because it’s ours and because it exposes the most complete tool set.

At least one social account connected inside the scheduler. The MCP server doesn’t authenticate with Instagram or LinkedIn directly. It talks to SchedPilot, and SchedPilot talks to the platforms through their official APIs. Connect your accounts in the dashboard first.

An API key. In SchedPilot, that’s under API Access in the app. It starts with smm_. API and MCP access are on the Platinum plan, which has a trial.

Node.js 18 or newer if you’re running the server locally through npx. Check with node -v. If you use the remote URL instead (below), you don’t need Node at all.

Setting it up in Claude Desktop

1. Open Settings → Developer → Edit Config. This opens claude_desktop_config.json.
2. Paste this in, replacing the token with yours:

{
  "mcpServers": {
    "schedpilot": {
      "command": "npx",
      "args": ["-y", "schedpilot-mcp"],
      "env": { "SCHEDPILOT_TOKEN": "smm_your_key" }
    }
  }
}

3. Quit Claude Desktop completely (Cmd+Q on Mac, not just closing the window) and reopen it.
4. Click the tools icon in the chat bar. You should see SchedPilot’s tools listed.

If the icon doesn’t appear, the JSON is almost always the problem. A trailing comma after the last brace will silently break the whole config. Paste it into any JSON validator and it’ll point at the line.

Setting it up in Claude Code

One command:

claude mcp add schedpilot \
  --env SCHEDPILOT_TOKEN=smm_your_key \
  -- npx -y schedpilot-mcp

Run /mcp inside a Claude Code session to confirm it’s connected. This is the setup I’d recommend for anyone who lives in a terminal anyway, because you can chain it with other tools. A Claude Code session that reads your changelog from a repo and schedules the release announcement to X and LinkedIn is a two-minute job once both servers are connected.

Setting it up in ChatGPT, Cursor, or anything with a remote URL

Most clients now accept a remote MCP endpoint instead of a local process. For SchedPilot, the endpoint is:

https://api.schedpilot.com/mcp

In ChatGPT: Settings → Connectors → Add custom connector, paste the URL, and authenticate with your SchedPilot key when prompted. In Cursor: Settings → MCP → Add server, choose the URL type, paste. Claude.ai’s web app also accepts remote connectors under Settings → Connectors, which means you can use SchedPilot from the browser without installing Claude Desktop.

The remote route is the one to pick if you don’t want Node on your machine or you’re setting this up for a teammate who won’t edit config files. It’s the same server and the same tools either way.

What the server actually lets the AI do

SchedPilot’s MCP server exposes seven tools. Knowing them helps you phrase requests the AI can act on.

get_accounts lists your connected social accounts and their IDs. The AI calls this first on almost every request, because it needs to know which Instagram account “my Instagram” refers to.

create_post schedules a post to one or more accounts. It takes the text, the accounts, the time, and optional media. This is the one that does the scheduling.

upload_media and list_media handle images and video. Upload from a public URL, or reuse something already in your library.

list_posts shows what’s scheduled, published, or failed. Useful for “what’s going out tomorrow” and “did the Thursday post publish.”

delete_post cancels a scheduled post.

get_analytics pulls impressions, reach and engagement for a published post.

There’s no edit_post. If a scheduled post needs a change, the AI deletes and recreates it. That’s a deliberate choice; it keeps the tool surface small enough that the model rarely picks the wrong one.

Scheduling your first post

Once the server shows up in your client, test with something concrete:

List my connected accounts, then schedule this to LinkedIn for tomorrow at 9am: “We just shipped bulk CSV scheduling. Here’s what changed.”

You’ll see the AI call get_accounts, then create_post, then tell you it’s scheduled. Open your SchedPilot calendar and it’s there.

Two things to check on this first run. Confirm the time zone: the AI interprets “9am” using the time zone SchedPilot has on your account, and if that’s wrong the post goes out at the wrong hour. And confirm which account it picked if you have more than one on the same platform. Both are easy to fix in the calendar, but it’s better to catch them once than to trust every future post blindly.

Prompts that work well

The more useful patterns aren’t single posts. They’re the ones where the AI does the tedious part.

Turning long content into a scheduled batch. Paste a blog post and say: “Turn this into five LinkedIn posts, one per key point, and schedule them for 9am on the next five weekdays. Don’t use hashtags.” The AI writes all five, calls create_post five times, and you review them in the calendar. This is thirty minutes of work reduced to two, and it’s the single most common thing people use the connection for.

Cross-posting with per-platform edits. “Schedule this announcement to X, LinkedIn and Threads for 2pm Thursday. Keep the X version under 280 characters and make the LinkedIn version more formal.” One request, three tailored posts.

Reading before writing. “What were my three best-performing posts last month? Write two new posts in the same style and schedule them for next Tuesday and Thursday.” The AI calls list_posts, then get_analytics on each, then create_post. Analytics informing new content without you opening a dashboard.

Filling gaps. “Show me what’s scheduled for next week. If any weekday has nothing on Instagram, draft something from my product FAQ and schedule it for 11am that day.” The AI reads the calendar and fills holes.

Killing something. “Delete the LinkedIn post scheduled for Friday about the pricing change.” Faster than finding it in a calendar.

We collected a longer list in 100 prompts to manage social media with AI, most of which work directly through MCP.

Where MCP beats the other automation routes

You could do a lot of this with n8n or Make. We have a guide to automating social posting with n8n and it’s a good fit for anything that runs on a fixed trigger, like “every time a blog post publishes, schedule a tweet.”

MCP is better for the unscheduled, judgment-heavy work. “Look at what’s queued and tell me if anything’s too salesy” isn’t a workflow you’d build in n8n. It’s a conversation. The AI reads the calendar, forms an opinion, and only writes when you tell it to.

The REST API is the third route, and the right one if you’re building a product rather than working in a chat. MCP sits on top of the same API. If you find yourself wanting to schedule from a script instead of a conversation, the API docs cover the same endpoints.

The approval question

Every post the AI creates through SchedPilot’s MCP server lands in your calendar as a scheduled post. It will publish at the set time unless you intervene, so if you want a human gate, set the default post status to pending approval in your workspace settings. Then the AI drafts and schedules, someone on your team approves, and nothing goes out on the strength of a prompt alone.

For a solo account, I’d skip the approval step and just review the calendar before bed. For an agency running client accounts, turn it on. An AI that confidently schedules a post to the wrong client account is a story you only want to tell once. Our comparison of approval workflow tools covers how different schedulers handle this.

Publishing itself goes through each platform’s official API, the same way it would if you’d clicked Schedule in the dashboard. There’s no browser automation, no unofficial endpoints, and nothing that puts the account at risk.

Troubleshooting

The tools icon never appears in Claude Desktop. Fully quit the app, check the JSON, and make sure node -v returns 18 or higher. Those three fix nearly every case.

The AI says it scheduled a post but nothing shows in the calendar. Usually the token. Re-copy it from API Access; a truncated key authenticates far enough to list tools but fails on writes.

create_post fails on Instagram with a media error. Instagram requires media on every post and rejects anything that isn’t 9:16 for Reels or one of its accepted feed ratios. Text-only “posts” to Instagram will always fail. Upload the media first, then reference it.

The AI schedules to the wrong account. Add the account handle to your prompt (“schedule to @schedpilot on X, not the personal account”). Once it’s picked correctly, later requests in the same conversation tend to follow.

The post went out at the wrong hour. Time zone mismatch between SchedPilot and what you meant. Set the workspace time zone once and the problem goes away.

Frequently asked questions

What is a social media MCP server? It’s a small service that exposes a scheduler’s actions (list accounts, create a post, upload media, read analytics) as tools an AI model can call. You connect it to Claude, ChatGPT or another MCP client, and the AI can schedule and manage posts on your behalf instead of only drafting text. SchedPilot’s server talks to SchedPilot’s API, which in turn publishes through each platform’s official API.

Which social media tools have MCP servers in 2026? SchedPilot has a first-party server (hosted and as an npm package). SocialPilot launched one in 2026. API-first services like Ayrshare can be wrapped in community-built servers, though those aren’t maintained by the vendor. Buffer, Hootsuite, Later, Sprout Social and Planable don’t offer MCP as of this writing. The list changes every few months, so check the vendor’s docs rather than a third-party roundup.

Is the MCP server included in every SchedPilot plan? No. API and MCP access are part of the Platinum plan. Lower tiers include the scheduler, calendar and analytics but not programmatic access. There’s a seven-day trial on Platinum, so you can test the MCP connection before paying.

Should I use the hosted server or run the npm package locally? Use the hosted endpoint (https://api.schedpilot.com/mcp) unless you have a reason not to. It needs no Node install, works from claude.ai in the browser, ChatGPT and Cursor, and it’s the easiest to hand to a teammate. Run the npm package (npx -y schedpilot-mcp) if you’re in Claude Code or Claude Desktop already and prefer everything local, or if your security policy requires the token to stay in a config file on your machine rather than in a client’s connector settings. The tools are identical either way.

What can an assistant actually do through the MCP server? Seven things: list your connected accounts, create and schedule a post to one or more of them, upload media from a URL, list media already in your library, list scheduled and published posts, delete a scheduled post, and read a published post’s analytics. It cannot connect new social accounts, change billing, or edit a post in place (it deletes and recreates instead). Everything it creates shows up in your SchedPilot calendar, so you can review or cancel before the post goes live.

Does it work with ChatGPT as well as Claude? Yes. ChatGPT accepts remote MCP servers through custom connectors on its paid plans; add the hosted SchedPilot URL under Settings → Connectors and authenticate with your API key. The tools and prompts are the same as in Claude. In practice Claude handles multi-step requests (read analytics, then draft, then schedule) a bit more reliably, but for straightforward scheduling both work. Cursor, Windsurf and Claude Code connect the same way.

Start with one week

Connect the server, paste in a piece of content you’ve already written, and ask for a week of posts across two platforms. Review them in the calendar. If the drafts need heavy editing, tighten the prompt with an example of your voice and try again. If they’re close, you’ve just recovered a few hours a week.

SchedPilot’s MCP server works on the Platinum plan, which you can trial for seven days with all nine networks connected: Instagram, X, LinkedIn, TikTok, Threads, Facebook, YouTube, Pinterest and Bluesky. If you’re a Claude user specifically, the step-by-step Claude guide has screenshots of every screen; the ChatGPT equivalent is in how to schedule tweets with ChatGPT.