ChatGPT social media automation: what it can post, and what it needs to post it

ChatGPT cannot post to Instagram or LinkedIn on its own. It needs a connector: an MCP server that holds your accounts and their tokens. Setup takes about five minutes in developer mode, and OpenAI still asks you to confirm every write action. Here is the exact flow, checked on October 12, 2026.

What ChatGPT does and does not do on social media

Out of the box, ChatGPT is a writer and an analyst. It has no account of yours, no token, no publish endpoint. Browsing a public profile is not posting to it.

The last row is the one people discover late, and it is the honest limit of the whole category. A chat client acts when someone is in the conversation. For publishing that happens without a human in the loop, you want a cron job on a REST API or an always-on agent, not a chat window.

Can ChatGPT post to Instagram?

Only through a connector, and the reason is Instagram, not OpenAI. Publishing to an Instagram business account means the Graph API, an app with the right permissions, a long-lived token you refresh every 60 days, and a media container you create then publish. None of that is something a chat model can hold. Tokens, refresh and the 60-day clock are covered in social media access tokens.

So the architecture is always the same three parts: ChatGPT decides and writes, an MCP server exposes the actions, a platform API executes. Swap ChatGPT for Claude and nothing else changes. That is also why I keep saying the model is the least interesting part of the stack.

Set up a ChatGPT MCP connector

Using PlugKit as the server, since it is what I run. PlugKit's client documentation has the current steps, and here they are in order:

1. Turn on developer mode in ChatGPT: Settings, then Security and login, then Developer mode. It is not under Apps or Connectors, which is where everyone looks first. You need the web app and a paid plan (Plus, Pro, Business, Enterprise or Edu).
2. Go to chatgpt.com/plugins, click + at the top right, then Create app.
3. In the New plugin dialog, ignore the drag-and-drop box and click Create an MCP app at the bottom left. Create plugin and Upload plugin never ask for a server URL, which is the dead end most people hit.
4. Name it PlugKit, add any description, and paste the server URL https://api.plugkit.co/mcp. Choose OAuth and leave client ID and secret empty.
5. Approve on the PlugKit consent screen. You pick read-only or full access there.
6. In a new chat, open + and select PlugKit. Test with "search my PlugKit account for the conversation about pricing". ChatGPT should call search, then fetch.

Transport is Streamable HTTP. Legacy SSE is not supported, and there is nothing to install locally since the server is remote.

If step 5 fails with a 401 later, disconnect and reconnect to rerun consent. And note what read-only costs you: a session opened with the mcp:read scope does not register the write tools at all, so ChatGPT will say it cannot publish. Reconnect with full access when you want it to post.

Why a ChatGPT connector needs search and fetch

This is the part that surprises developers building their own server. OpenAI's MCP documentation says a server should "implement two read-only tools: search and fetch" for ChatGPT's deep research and company knowledge surfaces. search takes a query and returns results with an id, a title and a url. fetch takes a document id and returns id, title, text, url and optional metadata. Results without a non-empty url are not citable and stay as plain tool output.

So ChatGPT's connector model is retrieval-shaped first and action-shaped second. A social MCP server that only exposes create_post and send_message will connect, but ChatGPT's discovery surfaces will have nothing to call. PlugKit ships search and fetch as read-only tools next to the publishing ones for that reason.

Two more rules from the same page, both worth reading before you trust a connector you did not write: ChatGPT requires manual confirmation before write actions, and custom MCP servers are not developed or verified by OpenAI. The first is a feature. The second means the consent screen is your own responsibility.

What you can actually automate

Sort your work by whether a human is present, and the tool picks itself.

The pattern: ChatGPT is excellent for the half of the job that needs judgment with you in the room. The unattended half belongs to a scheduler or an agent process. I use both, and the deciding question is never which model, it is whether anyone is awake.

For the same walkthrough on the other side of the market, see Claude as your social media manager. For the protocol choice underneath, MCP or REST API for your agent. For what a server like this exposes per platform, the Instagram MCP server page goes tool by tool, and social media MCP servers compared is the honest field guide.

What it costs

ChatGPT's side: a paid plan, because developer mode and custom MCP apps require a paid subscription. PlugKit's side: $49 a month for 5 connected accounts, cancel anytime, or $97 for 30 with extra accounts at $3. There is a 7-day free trial, no card, with one connected account, and the clock starts when you connect it. X API calls are passed through at cost, $0.005 per read and $0.015 per publish, because X charges per call.

Where ChatGPT is the wrong choice: if your flow is "same post, every day, no decisions", you are paying a model to do a cron job's work, and a 30-line script calling the REST API will be cheaper, faster and testable. If you want the connector mainly for analytics reading, a dashboard is less work. ChatGPT earns its place when the content changes every time and you want to argue with it before publishing.

FAQ

Can ChatGPT post to Instagram directly?

No. ChatGPT has no access to your Instagram account and Instagram requires the Graph API with a business or creator account. Connect an MCP server that holds the account, such as PlugKit, and ChatGPT can then call a publish tool. It will ask you to confirm the write first, because OpenAI requires manual confirmation before write actions.

How do I add a social media MCP connector to ChatGPT?

Turn on developer mode under Settings, Security and login. Open chatgpt.com/plugins, click +, then Create app, then Create an MCP app. Paste the server URL, choose OAuth, approve the consent screen, then pick the connector with + in a new chat. A paid ChatGPT plan and the web app are required.

Can ChatGPT run social media automation while I am away?

Not reliably, and not for writes. A chat client runs when you prompt it, and ChatGPT confirms write actions with you before executing them. For scheduled publishing or instant DM replies, use a REST API on a scheduler, a platform automation, or an always-on agent process.

Is ChatGPT or Claude better for this?

Neither, for this purpose. Both connect to the same MCP server over the same transport and call the same tools. Claude allows custom connectors on all of its tiers, limited to one connector at the entry level, while ChatGPT's developer mode needs a paid plan, so the practical difference is which subscription you already pay for.

Do I need to write my own MCP server?

Only if no existing server covers your platforms. If you do write one, implement search and fetch as read-only tools so ChatGPT's discovery surfaces have something to call, return a non-empty url in results so answers can cite them, and expect no verification from OpenAI: your consent screen and scopes are yours to get right.

AB

Antoine BlancoFounder of PlugKit. Runs his own Instagram, YouTube and X accounts through the API and MCP server described here.