How to schedule social media posts with ChatGPT

ChatGPT writes a decent first draft of a social post. The slow part is everything after: copying it into a scheduler, cutting it for X, finding an image for Instagram, picking a time. Connecting ChatGPT to a scheduling platform over MCP removes that copy-paste step.

To schedule social media posts with ChatGPT, you add a social media MCP server as a custom connection, sign in once, and ask ChatGPT to draft and schedule. The platform makes a version per network, checks your brand rules, and holds the post for a person to approve. As of November 2026, OpenAI limits these write actions to Business, Enterprise and Edu workspaces.

This guide uses Tracee as the example server. The ChatGPT steps are the same for any remote MCP server.

Which ChatGPT plans can schedule posts?

As of November 2026, only ChatGPT Business, Enterprise and Edu get full MCP support, which includes write actions such as creating a draft or scheduling a post. Pro can connect custom MCP servers with read and fetch permissions only. These details come from OpenAI's help article on connecting custom MCP servers in ChatGPT, updated in late October 2026.

Two more limits from the same article: you can't connect a custom MCP server from the mobile app (web only), and deep research can use custom apps for reading only.

You also need a Tracee workspace with at least one social account connected. Every Tracee plan includes MCP access, starting at $10 a month at the time of writing (November 2026), and the trial runs 7 days without a credit card. See Tracee plans.

If you are on Pro, the connection still answers questions like "what did we post last week and how did it do?". For drafting and scheduling, the last section of this guide lists what to do when your plan can't write.

How do you connect Tracee to ChatGPT?

You add Tracee from the Plugins page on chatgpt.com. In the current Plugins experience you don't need to turn on Developer mode first; you acknowledge a risk warning instead. These steps follow OpenAI's connect and test guide as of November 2026.

1. Open chatgpt.com/plugins on the web.
2. Select the plus button, then Add custom MCP server.
3. Enter a name (Tracee) and, if you like, a description such as "Drafts and schedules our social posts".
4. Under Connection, enter https://traceeai.com/mcp.
5. For authentication, choose OAuth. Tracee supports OAuth 2.1 with dynamic client registration.
6. Read the risk warning and tick I understand and want to continue.
7. Select Create as a plugin, then review the tools ChatGPT discovered (posts, channels, brand, approvals, analytics and the rest).
8. Install the plugin and sign in to Tracee when ChatGPT asks. If your account is in more than one Tracee workspace, tell ChatGPT which one to use.

On Business, only admins and owners can create the connection and publish it for the workspace; the workspace permission is called Create plugins with MCPs. After an admin publishes it, members find it under Settings > Plugins.

Some workspaces still show the older flow. There, admins go to Workspace settings > Apps > Create, and authorized users go to Apps > Create in their own settings, enter the URL, pick OAuth, select Scan Tools and then Create. Older guides, including earlier versions of ours, describe a Developer mode toggle; follow whichever screen your workspace shows.

Want ChatGPT to only write drafts?

Tracee also serves a drafts-only endpoint built for ChatGPT's app directory: https://traceeai.com/mcp/chatgpt. Through it, ChatGPT can read your workspace and create drafts, but it can't schedule, publish, send or delete. A person schedules from Tracee. Use this URL in step 4 if that is the policy you want.

How do you write the first prompt?

Start with a read-only prompt, so you can see the connection work before anything is written. In ChatGPT, type @, pick the Tracee plugin, then ask. OpenAI notes that the selection applies to that message, so mention the plugin again when a follow-up needs new data.

A good first prompt:

@Tracee Which channels are connected, and what is scheduled for the next 7 days?

Then a real request. Give ChatGPT the facts it can't guess: the link, the audience, the one point that matters, and the time.

@Tracee Write a post announcing our new reporting dashboard (https://example.com/launch) for LinkedIn and X. Audience: B2B marketing leads. Main point: they can see which posts brought in contacts. Read our brand rules first. Schedule it for Tuesday at 9:00 our time.

Behind the scenes, ChatGPT calls brand to read your tone and banned phrases, channels to find the LinkedIn and X accounts, posts with action: draft to create one version per channel, and posts with action: schedule. The link goes out with UTM tags per channel, so a contact who clicks can later be traced back to this post in your CRM.

How do you review drafts before they go out?

Review every network's version, including the short X one and any Instagram one. Tracee turns one draft into a version per channel and holds each to that network's rules: the X version stays within 280 characters, an Instagram version needs an image or video, and TikTok and YouTube need a video. If a version breaks a brand rule, such as a banned phrase, it is flagged and can't go out until fixed.

You can review in two places:

• In ChatGPT: ask "Show me each version and any brand issues." Then edit by asking: "Make the X version shorter and drop the second sentence."
• In Tracee: the post sits in the calendar with its versions side by side, and you can edit the copy directly.

ChatGPT may also ask you to confirm a write action before running it. OpenAI says ChatGPT decides this based on the app's permissions and the action's context, so you will see more prompts for scheduling than for reading.

How do approval and scheduling work?

With approval on, which is Tracee's default for agents, ChatGPT's schedule request becomes an approval request. The answer comes back as pending_approval, and ChatGPT should tell you the post is waiting for a person.

1. A teammate opens Tracee and sees the post in the Approvals queue.
2. They approve it, and it goes out at the time ChatGPT asked for. Pressing Publish in the app also counts as approval.
3. If they reject it, nothing goes out, and ChatGPT can see the decision the next time you ask.

ChatGPT can't approve its own request; no tool does that. You can check from the chat with "Did Tuesday's post get approved?", which calls approvals with action: list.

A person can turn off approval for agent posts in Settings > Agent rules, and limit which channels agents may post to. Deleting a post or taking one down always waits for a person, whatever that setting says. For a fuller policy, see our guide to an approval workflow for AI agents.

Which prompts work well?

Prompts that name the channel, the time and the source material work best. These are tested patterns:

Troubleshooting

What are the limits?

• Plans: write actions need Business, Enterprise or Edu as of November 2026.
• Mobile: setup is web only.
• Networks: an agent can't post a format the network's API doesn't allow. Tracee publishes to 13 networks, from LinkedIn and Instagram to Reddit and Google Business Profile.
• Speed: each Tracee credential gets 120 requests a minute, which is far above what a chat uses.
• Judgment: ChatGPT writes what you ask for. A vague request ("post about the launch") gets a vague post.

What if your ChatGPT plan can't write?

You have three options that keep the same workflow:

1. Use Claude. As of November 2026, Anthropic allows custom connectors on every Claude plan, including one on Free (Claude help center). Our guide to scheduling posts with Claude has the steps.
2. Use Tracey, the assistant built into Tracee. It does the same drafting and scheduling work inside the app.
3. Build on the OpenAI API. Agents built on the Responses API or the Agents SDK can connect to https://traceeai.com/mcp with a Tracee workspace key as a bearer token. The MCP setup docs show the tool config.

If MCP itself is new to you, start with our explainer on what an MCP server for social media is.

Next step

Connect the server, run one read-only prompt, then schedule one low-stakes post with approval on and watch it move from ChatGPT to the Approvals queue to the network. Once that loop works, give ChatGPT a week of source material and ask for the whole week's drafts in one go.