DEV Community: Olver Thorarinsson

Which AI Can Actually Schedule Your Posts Over MCP? ChatGPT, Claude, Cursor, Gemini CLI and VS Code (2026)

Olver Thorarinsson — Mon, 19 Oct 2026 19:10:27 +0000

Originally published at smartpoststudio.com. Republished with a canonical link.

Not every AI assistant that "supports MCP" can actually schedule a post. ChatGPT is the big surprise: OpenAI's own help center says full MCP, the kind that can create or change things, is limited to Business, Enterprise and Edu workspaces, and on Pro a custom MCP connection gets read and fetch permissions only. That means it can tell you what's in your queue, but it can't draft or schedule. Below is the client-by-client picture for October 2026: what each assistant can do with a scheduler, how you connect it, and which ones we've documented ourselves.

A note on how to read this. Everything about third-party clients comes from each vendor's own documentation, checked on October 12, 2026, and these features change monthly. Where we say a client "should work" with Smart Post Studio, that means its documented connection method matches ours, not that we've published a tested walkthrough for it. We'll say which is which.

The three questions that decide whether a client can schedule

Before the list, here's what actually matters. MCP itself is an open standard, but clients implement different slices of it.

1. Can it reach a remote server? A scheduler's MCP server lives on the internet (ours is at https://app.smartpoststudio.com/v1/mcp). Some clients only run local servers on your own machine. All five below can reach remote ones.

2. How does it sign in? Either OAuth (a browser window opens, you log in, you pick permissions) or a token pasted into a config file as a header. OAuth is friendlier; tokens work in more places.

3. Will it call write tools? Reading your queue is a read. Creating a draft, scheduling, moving or cancelling a post are writes. This is the column that catches people out.

Claude (claude.ai and the desktop app)

Can schedule: yes. Anthropic lists custom connectors using remote MCP on the Free, Pro, Max, Team and Enterprise plans, with Free accounts limited to one custom connector. On Team and Enterprise, an owner adds the connector for the organization first and members then connect individually.

How: Settings → Connectors → Add custom connector, paste the server URL, leave the OAuth client ID and secret empty, sign in, then choose permissions. This is the setup we've documented step by step on our MCP page and in the Claude setup guide.

One thing to be clear about: a free Claude account can hold the connector, but our MCP access itself is a paid Smart Post Studio feature (from the $8/month Scheduler plan). The free side of the equation is Claude's, not ours.

Claude Code

Can schedule: yes. It's the same OAuth sign-in, started from the terminal:

claude mcp add sps --transport http https://app.smartpoststudio.com/v1/mcp

Then run /mcp and finish the sign-in in your browser. Useful if you already live in a terminal, or want to script a weekly routine around the prompts in our seven-prompt guide.

Cursor

Can schedule: yes. Cursor's docs list stdio, SSE and Streamable HTTP transports, with OAuth for the remote ones, and you add servers in mcp.json (per project or globally). Cursor also supports static headers for servers that use tokens. Cursor is one of the clients we name on our MCP page: OAuth clients sign in by adding the URL, and header-based setups use a token created under web app → AI access.

ChatGPT

Can schedule: only on Business, Enterprise or Edu, according to OpenAI's help center. Here's the detail, because the two OpenAI pages don't read the same.

OpenAI's developer documentation describes developer mode as full MCP client support for read and write tools, available to Pro, Plus, Business, Enterprise and Education accounts on the web. You switch it on under Settings → Apps → Advanced settings → Developer mode, then use "Create app" to point at a remote server. It supports SSE and streaming HTTP, and OAuth or no authentication. There's no header-token option in that list.

The help center article, updated in October 2026, is narrower. It says full MCP support, including modify and write actions, is rolling out in beta to Business, Enterprise and Edu, and that Pro users can connect MCP servers with read and fetch permissions in developer mode. It doesn't spell out Plus separately. Until you've tested it on your own account, we'd assume an individual plan can read your queue but not change it.

Other things worth knowing from the same pages:

MCP apps are web-only. They aren't available in the mobile app.
Write actions ask for confirmation by default. You can tell it to remember an approval for the rest of a conversation, and OpenAI recommends only doing that for apps you trust.
On Business, only admins and owners can use developer mode. On Enterprise and Edu, admins can grant it to named people.
ChatGPT's agent mode won't use custom apps. Deep research can, but only for reads.

Where that leaves Smart Post Studio: ChatGPT's OAuth flow should be able to reach our server, since it's the same remote URL Claude uses, but we haven't published a verified ChatGPT walkthrough, and on most individual plans you'd be limited to questions like "what's scheduled this week?" anyway. If you're on a Business workspace and get it working, we'd like to hear how it went.

VS Code with GitHub Copilot

Can schedule: should work (we haven't documented it). VS Code stores MCP servers in mcp.json, either in your workspace (.vscode/mcp.json) or your user profile. A remote server needs a type of http and a url, and the reference documents a headers field plus an inputs array so you're prompted for a token rather than storing it in the file. Using a token from web app → AI access, that looks like:

{ "servers": { "sps": { "type": "http", "url": "https://app.smartpoststudio.com/v1/mcp", "headers": { "Authorization": "Bearer ${input:sps-token}" } } }, "inputs": [ { "id": "sps-token", "type": "promptString", "description": "Smart Post Studio token", "password": true } ] }

If you're on Copilot Business or Enterprise, your organization's "MCP servers in Copilot" policy has to be enabled first. That's the most common reason a remote server silently doesn't appear.

Gemini CLI

Can schedule: should work (we haven't documented it). Gemini CLI reads MCP servers from ~/.gemini/settings.json (or a project's .gemini/settings.json). Remote servers go in with an httpUrl, and its docs describe OAuth discovery: if the server answers with a 401, the CLI discovers the sign-in details, opens a browser, and stores the token locally. You can trigger it with /mcp auth <server-name>. For servers that use tokens instead, a headers entry works the same way as in the other clients.

The short version, as a table

Client	Transport	Sign-in	Read + write?	Status
Claude (Free/Pro/Max/Team/Enterprise)	remote MCP	OAuth	yes	Documented by us
Claude Code	remote MCP	OAuth	yes	Documented by us
Cursor	remote MCP	OAuth or token headers	yes	Named on our MCP page
ChatGPT Business / Enterprise / Edu	remote MCP, web only	OAuth	yes, with confirmation	Not yet documented by us
ChatGPT Pro (and likely Plus)	remote MCP	OAuth	read and fetch only	Can check your queue, can't schedule
VS Code + Copilot / Gemini CLI	remote MCP	OAuth or token headers	should work	Not documented by us

Whichever client you pick, the guardrails live on the server

This is the part that doesn't change between clients. A confirmation dialog in ChatGPT, or permission checkboxes in Claude, are the client's safety features. What the assistant is able to do is decided by the scheduler's server. In Smart Post Studio's case: there's no "post now" tool for any client, anything an assistant schedules must be at least ten minutes out, it shows in your Queue with an AI badge, and it's written to your account history. There's a cap of 100 AI-created posts a day and 30 requests a minute. Each connection appears under AI access, lasts 90 days, and stops working the moment you delete it.

That ten-minute window is a chance to intervene, not an approval step. A scheduled post goes out at its time whether you looked at it or not. If you want the background on why we built it that way, MCP for people who don't code covers it without the jargon.

And if you're still choosing a scheduler rather than a client, our comparison of schedulers that speak MCP is honest about who's cheaper. Buffer, for one, lists Claude, ChatGPT and Cursor integrations on its free plan. The 2026 scheduler pricing study has the rest of the numbers.

TL;DR: Claude (including a free account), Claude Code and Cursor can all read and schedule over MCP today, and they're the clients we've documented. ChatGPT can too, but OpenAI's help center limits write access to Business, Enterprise and Edu. On Pro it's read and fetch only, so it can check your queue but not fill it. VS Code with Copilot and Gemini CLI both support remote servers with OAuth or token headers and should work, though we haven't published a walkthrough. Whatever the client, the scheduler's server sets the real limits.

Config files, copy-paste ready: every client setup in this post (Claude, Claude Code, Cursor, VS Code, Gemini CLI) lives as a file in our public GitHub repo, alongside a week of example prompts.

7 Claude Prompts That Run a Week of Social Media Over MCP (Copy-Paste Ready)

Olver Thorarinsson — Mon, 19 Oct 2026 19:00:40 +0000

Originally published at smartpoststudio.com. Republished with a canonical link.

Connecting Claude to a scheduler over MCP takes about a minute. Knowing what to actually say to it afterwards is the part nobody writes down, so most people ask "what's in my queue?", get an answer, and never come back. Below are the seven prompts we'd hand a solo creator on day one, in the order you'd use them across a week, each with what it does, why it's worded that way, and where it stops.

A quick note on scope. These are written for Smart Post Studio's MCP connector because that's the one we can describe precisely, limits included. Most of the wording carries over to any scheduler with an MCP server; the guardrails won't, so check what yours permits before you paste prompt five.

Before the prompts: set the permissions you'll actually use

When you connect, you choose what the assistant may do: read (see your queue and results), draft (write drafts for your review) and schedule (put posts on the calendar). Start with read and draft. Add schedule after a week of drafts you didn't need to rewrite, not before.

One thing doesn't change whichever boxes you tick: there is no "post now" tool. Anything an assistant schedules has to be at least ten minutes out, appears in your Queue with an AI badge, and is written to your account history. That's a window to intervene, not an approval step. A scheduled post goes out at its time whether or not you looked at it. The full setup is on our MCP page, and the step-by-step version is here.

Prompt 1 — Monday: get the lay of the land

"What's scheduled across all my platforms for the next seven days? Group it by day, flag any day with nothing on it, and tell me which platforms each post goes to."

This is read-only, so it's safe to run with the narrowest permissions. The useful part isn't the list. It's "flag any day with nothing on it", because gaps are what you're actually looking for on a Monday. Asking for the platform per post also surfaces the post you meant for Instagram and Threads that's only going to Instagram.

Prompt 2 — Find out what you can reach

"Which platforms can I post to right now, and which of them publish from the cloud versus from my desktop app?"

Run it once, and again whenever you connect something new. The assistant can only post to accounts you've already connected; it can never add one itself. The cloud/desktop split matters for one rule: a single post can't mix cloud-published and desktop-published platforms. If you ask for one that does, the assistant is told to split it in two, and it helps to know why that's happening before it does.

Prompt 3 — Turn one idea into platform-shaped drafts

"Draft posts about [the thing] for Instagram, Threads and LinkedIn. Same idea, different openings: Instagram leads with the visual, Threads reads like a reply to a friend, LinkedIn leads with the lesson. Keep Threads under 500 characters. Save them all as drafts — don't schedule anything."

Three things are doing work here. "Different openings" stops the copy-paste effect that makes cross-posting obvious (we went deeper on that in this guide). The explicit character cap matters because models are bad at counting; if a limit is critical, check it yourself in our character counter before it ships. And "don't schedule anything" is belt-and-braces: even with schedule permission granted, you want this step to end in your drafts.

If you have an image or video, include a link to it in the prompt. Media goes in by link, up to 25 MB per file.

Prompt 4 — Edit in plain language

"In the LinkedIn draft, cut the first sentence, replace 'game-changer' with something a person would say, and end on a question. Show me the new version before saving it."

"Show me before saving" is the habit worth building. You're reviewing in the chat where it's cheap, rather than in the queue after the fact. If the edits keep being the same kind of edit, that's a signal to put your style rules at the top of every session instead. We wrote about which rules actually hold in using AI without sounding like it.

Prompt 5 — Schedule with the time zone spelled out

"Schedule the Instagram and Threads drafts for Tuesday at 10:00 a.m. Eastern, and the LinkedIn one for Wednesday at 8:30 a.m. Eastern. Confirm each time back to me in Eastern before you do it."

Always name the time zone. "Tuesday at ten" is ambiguous to a model in a way it isn't to you, and the confirm-back step catches the mistake before it's on your calendar. If your audience is spread across zones, work out the overlap first. Our post time planner does it without a signup. This prompt also needs schedule permission; it fails cleanly without it, which is the point of granting it late.

Prompt 6 — Move and cancel without opening the app

"Something came up. Move everything scheduled for Thursday to Friday at the same times, and cancel the X post about the sale entirely."

This is where conversational scheduling earns its keep over a calendar UI. Rescheduling a whole day is one sentence. "Cancel" returns the post to your drafts rather than deleting it, so you don't lose the copy if the sale comes back.

Prompt 7 — Friday: check what actually went out

"For every post scheduled this week, did it publish? List anything that failed, on which platform, and the reason given."

This closes the loop. What comes back is the per-platform publish result. It isn't an analytics report: it tells you whether a post went out, not how it did. For performance, use each platform's own insights, or plug the numbers into our engagement calculator. One common reason a desktop-published post didn't go out: the desktop app wasn't running at that time. Cloud-published platforms don't have that dependency.

Three prompts we'd avoid

"Run my social media this week." Too open. You'll get volume, not judgement, and you'll spend longer reviewing it than writing three good posts yourself.

"Post this now." It can't. The earliest anything goes out is ten minutes from the request, and that's deliberate.

"Schedule 50 posts for next month." It's technically possible within the limits (100 AI-created posts a day, 30 requests a minute), but bulk-generating a month in one go is how accounts start to sound like nobody. Draft in batches you'll actually read.

What this costs, and what it doesn't do

MCP access is included on every paid Smart Post Studio plan, from the $8/month Scheduler plan up. It isn't on the free plan. It's in beta: built, tested and live, but new. If you only want the cheapest possible way to point an AI at a scheduler, we're not it, and our MCP comparison says who is. What you get here is the ten-minute floor, the AI badge and the audit trail, which matter more the longer you leave an assistant running.

If you want to see what a month of this looks like in practice, including the part that went wrong, we wrote it up: we let Claude run a social account for a month.

TL;DR: Seven prompts cover a week: survey the queue (1), check reachable platforms (2), draft per platform without scheduling (3), edit with a preview (4), schedule with the time zone named and confirmed (5), move or cancel in one sentence (6), and check publish results on Friday (7). Start with read and draft permissions, add schedule last, and remember the assistant can't publish instantly. That's by design.

Config files, copy-paste ready: every client setup in this post (Claude, Claude Code, Cursor, VS Code, Gemini CLI) lives as a file in our public GitHub repo, alongside a week of example prompts.

Which Social Media Schedulers Support MCP? Every Tool Compared (2026)

Olver Thorarinsson — Mon, 19 Oct 2026 18:59:37 +0000

Originally published at smartpoststudio.com. Republished with a canonical link.

Six months ago, "my scheduler has an MCP server" would have been a real differentiator. In July 2026 it isn't — at least six schedulers now have one, and any tool implying otherwise is hoping you won't check.

So this isn't a post about who has MCP. It's about the four things that actually differ once nearly everyone does: what it costs, what your AI is allowed to do, where your media ends up, and which platforms it reaches.

We build one of these tools, so read this with appropriate suspicion. We come second in the first table and we're not going to hide that.

The one-minute setup

The step-by-step version, with prompts, limits and troubleshooting, is our guide on how to schedule social media posts with MCP. The short form:

The setup is nearly identical whichever tool you pick, and it takes about a minute:

Get the MCP URL from your scheduler's settings (in Smart Post Studio: Settings → MCP Access; other tools bury it under "integrations" or "API").
Add it as a connector in your AI client — in Claude: Settings → Connectors → Add custom connector, paste the URL, and approve the OAuth prompt that opens.
Ask for what you want in plain language: "Draft a week of posts about X and schedule them Tuesday/Thursday mornings." The AI reads your queue, drafts, and schedules — you review in the app as usual.

That's the whole trick. The real decision is which scheduler sits on the other end — which is what the rest of this post compares.

Which social media schedulers have MCP, as of October 2026

Tool	MCP	Cost of access	Can the AI publish instantly?
Buffer	Official, first-party, hosted	Free on every plan, including the free tier	Yes
Postiz	Official, open-source	Self-host from ~$5/mo infrastructure; cloud from ~$23/mo	Yes — explicitly
PostSyncer	Official	From $7/mo (10 accounts)	Drafts, schedules, replies, reads analytics
RecurPost	Official	From $9/mo	Yes, for Facebook Pages per its docs
Socialync	Official	Business tier, $40/mo	—
Hootsuite	Yes — MCP connectors for Perch (publishing), Nest (inbox) and Lumen (insights), announced 24 June 2026, documented at hootsuite.com/integrations/mcp (confirmed 14 Oct 2026)	Requires a paid Hootsuite plan for each product	Perch connector covers content planning and publishing
Sprout Social	Gated behind Advanced	$399+ per user / month	—
Later	None that we could find	—	—
Smart Post Studio	Yes, beta (29 July 2026)	Every paid plan, from $8/mo	No — by design

Three things are obvious from that table.

We are not first, and we are not cheapest. Buffer shipped a first-party MCP server before we did and gives it away on every plan including the free one, with native support across Claude, ChatGPT, Cursor, Raycast and Perplexity — their words, on their own developer page. Postiz is open-source and self-hostable for roughly the cost of a coffee. PostSyncer undercuts us at $7. If your criterion is "cheapest way to point an AI at a scheduler", we are not the answer and you should stop reading here.

Sprout Social's pricing is the outlier. Gating agent access behind a $399-per-user tier is aimed at organisations with procurement departments rather than people.

Almost everyone lets the AI publish immediately. That's the column worth staring at, and it's the next section.

What your AI is allowed to do

This is where the tools genuinely diverge, and most comparisons skip it entirely.

Postiz's MCP tool creates, drafts or immediately publishes. Buffer's server can publish. RecurPost's can schedule or post immediately. In each case, an AI agent holding a valid token can put words on your public account with no human between the model and the audience.

Ours cannot. There is no immediate-post tool in Smart Post Studio's MCP server at all. Every AI action produces either a draft or a scheduled post with at least ten minutes' lead time, sitting visibly in your queue, editable and cancellable, with every action written to your account's audit trail.

Whether that's a feature or a limitation depends entirely on what you're building.

If you want an autonomous pipeline running content from generation to live with nobody in the loop, our ten-minute floor is an obstacle, and Postiz or Buffer will serve you better. We're comfortable saying that.

If the idea of a model publishing to your account without you seeing it first makes you uneasy — and it should at least give you pause, because language models still hallucinate and tokens still leak — that floor is the reason to pick us. We took the view that unsupervised write access to someone's public voice is a poor default, and that ten minutes costs nothing worth having.

Reasonable people land in both camps. Just make sure you know which camp your tool chose on your behalf.

Where your media ends up

Every cloud scheduler works the same way: you upload your video to their servers, they store it, and at the scheduled time they upload it onward to the platform. Adding MCP changes nothing about that — it just means an AI triggers the upload instead of you.

Two consequences follow. Your original file sits on someone else's infrastructure, and it's re-encoded in transit, which is why cloud schedulers cap file sizes well below what the platforms themselves accept.

Smart Post Studio splits the job. Scheduling happens in the cloud, so your AI can reach it from anywhere. Publishing happens from your own machine, straight to the platform's API, at original quality. Your 4K file never touches our servers.

One honest caveat before we claim uniqueness: Postiz self-hosted gets you to a similar place by a different route. Run it on your own server and your media never reaches a third party either. The difference is that you're then running a server — Postgres, Redis, updates, backups — whereas ours publishes from the computer you already own and already turn on. Among hosted MCP schedulers we have not found another that keeps your original file off its own infrastructure — but to our knowledge is not the same as for certain, and we would rather state the weaker true thing than the stronger convenient one.

The honest cost: for desktop-published platforms, your computer needs to be running when the post fires. Platforms connected in the web app publish from our servers and carry no such requirement. That's a real trade-off, not a free win. If your machine is rarely on, use the web-connected route.

Which platforms it reaches

An MCP server is only as useful as the platforms behind it. Ours covers eight: Instagram, Facebook, Threads, TikTok, X, YouTube, LinkedIn and Fanvue.

On Fanvue specifically, we need to be precise, because it's easy to overclaim here. Fanvue has its own official MCP server — it can reply to fans, pull insights and manage posts. So "the only way to reach Fanvue from an AI" would be false.

What's accurate is narrower, and hedged on purpose: to our knowledge we are the only multi-platform scheduler whose MCP server reaches Fanvue. We have not found another; that is not the same as proving none exists. Fanvue's own server manages Fanvue; it won't cross-post your Reel to Instagram and Threads. If you want one assistant scheduling one idea across Fanvue and your public accounts together, that's the gap we fill. Fanvue works from the web app and the desktop app alike.

Where we lose on coverage: no Pinterest, Bluesky, Mastodon or Google Business Profile. Socialync covers Bluesky; Postiz advertises 30+ platforms on self-host. If your mix includes those, we don't reach them yet.

So which should you pick

You want free, today — Buffer. First-party MCP, free tier, widest client support.
You want to own the whole stack — Postiz. Open-source, self-hostable, 30+ platforms.
You want cheapest paid with many accounts — PostSyncer at $7 for 10 accounts.
You're an enterprise with procurement — Sprout Social, if that tier is already budgeted.
You want a window to catch what your AI wrote before it goes out — us. The ten-minute floor and audit trail are the entire point.
You post large video and care where the original lives — us, or self-hosted Postiz if you're happy running a server. We publish from the machine you already own; Postiz keeps it on a server you administer.
You post to Fanvue alongside public platforms — us, for the cross-posting. Fanvue's own MCP if Fanvue is all you need.
You need Pinterest, Bluesky or Google Business Profile — not us, yet.

MCP support becoming table stakes is good for everyone using these tools. It moves the interesting questions back where they belong: what the software does with your content, and what it permits a machine to do on your behalf.

Setup walkthrough for ours: how to schedule social posts by talking to Claude. Plans and what's included are on the pricing page.

How we checked

Because a comparison post is only worth the sourcing behind it: Buffer's claims here were read directly from Buffer's own developer page on 30 July 2026. Postiz, PostSyncer and Socialync state their MCP support on their own sites. RecurPost's appears in third-party server directories. Sprout Social's tier gating comes from third-party reporting we could not confirm on Sprout's own pricing page.

A general note on the negatives in that table: where we say a tool does not offer MCP, we mean we could not find evidence that it does. None of these companies publishes a list of things they have decided not to build, so absence of a claim is not proof of absence of a feature. Treat every "none" here as "none to our knowledge, on the date we checked".

Does Hootsuite offer MCP connectors?

Yes, as of 24 June 2026. When we first published this post Hootsuite's integrations page didn't mention it and we marked it unknown; it is now documented at hootsuite.com/integrations/mcp. Hootsuite ships three MCP servers, one per product: Perch (content creation, planning and publishing), Nest (social inbox and customer care) and Lumen (insights, from Talkwalker), orchestrated by its "Wisdom" AI layer. They connect to Claude and ChatGPT as custom connectors — same Settings → Connectors → Add custom connector flow as ours — and each requires the relevant paid Hootsuite product. The practical difference from a solo creator's point of view is price: Hootsuite's MCP access rides on plans that start at $99/month, versus $8/month here. Corrected 14 October 2026.

Config files, copy-paste ready: every client setup in this post (Claude, Claude Code, Cursor, VS Code, Gemini CLI) lives as a file in our public GitHub repo, alongside a week of example prompts.

MCP support is moving quickly and these details will drift. We re-check this post quarterly. If something here is out of date or wrong — including anything that flatters us unfairly — tell us and we will correct it and note that we did.