An MCP server for social media is a connector that lets an AI assistant (Claude, ChatGPT, Cursor, Claude Code and others) read your social media calendar and create, schedule or delete posts through a scheduling tool. MCP, the Model Context Protocol, is an open standard: the scheduler publishes a set of tools, the AI client discovers them, and you authorize the connection once. After that you can say "schedule this for Friday at 9:00 on LinkedIn" in a normal chat and the post appears in your calendar.
This guide explains how social media MCP works, which tools a good server exposes, how to connect each major AI client, and the safety checks worth doing before you let an agent touch your accounts.
Key takeaways
- MCP lets an AI use your scheduler's functions directly, instead of producing text you copy and paste.
- A remote MCP server with OAuth is the simplest setup: paste one URL, sign in, click Allow.
- The tools matter more than the brand: look for listing posts, creating posts in several modes (draft, approval, schedule, now) and deleting with confirmation.
- Approval-first behavior is the most important safety feature: AI-created posts should wait for a human.
- MCP carries text and parameters, not files in most setups, so plan for adding photos and videos in the scheduler itself.
What MCP is, in plain terms
Before MCP, connecting an AI to a product meant a custom plugin for each AI. MCP standardizes this. There are three parts:
- The client: the AI app you talk to, such as Claude, ChatGPT, Cursor or VS Code.
- The server: a service run by the product (here, the social media scheduler) that describes what it can do.
- The tools: named actions with parameters, such as
create_postwith text, channels, a date and a mode.
When you ask for something, the AI decides which tool to call, fills in the parameters, and the server runs it with the same rules as the product's own interface. If the server refuses (for example, the text is too long for X), the AI sees the error and can fix the post. There is a short definition in our glossary entry on MCP and a broader one on AI agents.
Why MCP matters for social media work
Social media is full of small, repetitive actions: check the calendar, write a post, adapt it per network, pick a time, repeat. MCP turns those into one sentence. Typical things people ask:
- "What is scheduled for next week in the Acme project? Only summarize."
- "Draft three LinkedIn posts about our launch and leave them for my approval on Monday, Wednesday and Friday at 9:00."
- "Turn this blog post into an X post and a Threads post, and schedule both for tomorrow at 18:30."
- "Delete the drafts I created yesterday."
MCP is not unique to one tool anymore. Several schedulers now offer official MCP servers, with different scopes: some focus on publishing, others add analytics or inbox features. Compare what the tools actually let the AI do rather than whether a server exists.
How a social media MCP server works: Postika as an example
Postika runs a remote MCP server at https://postika.app/api/mcp. It uses Streamable HTTP and OAuth 2.1, so compatible clients handle sign-in automatically: you paste the URL, the client opens a Postika window, you sign in and click Allow. OAuth connections reach every project of the user. For agents that do not support OAuth, a project API key works instead and reaches only that key's project.
The five tools
| Tool | What it does | Changes data? |
|---|---|---|
list_projects | Lists your projects with ID, name, time zone and current local time | No |
list_channels | Lists connected channels: network, handle, character limit, whether media is required | No |
list_posts | Lists calendar posts for a range of days (1 to 62, default 14) | No |
create_post | Creates a post on one or more channels | Yes |
delete_post | Deletes a post from Postika (drafts, pending, scheduled or published) | Yes, marked destructive |
The four posting modes
create_post takes the text, the channel IDs, a mode and, when needed, a time:
approval: the post waits in the calendar as pending approval until you approve it in the app.schedule: the post is scheduled for the given time.draft: saved as a draft, time optional.now: published immediately.
Dates are always in the project's local time, written as YYYY-MM-DD HH:MM. With a single project, the AI does not even need the project ID.
By default the server tells the AI to prefer approval unless you explicitly ask it to schedule or publish. That default is the difference between an assistant and an unsupervised bot.
What happens to photos and videos
Media cannot be sent over MCP. Instagram, TikTok, Pinterest and YouTube require a photo or video, so for those networks the AI creates a draft or pending post and you add the media in the app before approving. Facebook, X, LinkedIn and Threads accept text-only posts, so the AI can prepare those end to end.
Deleting a live post
Deleting a post that is already published removes it from your calendar only; it stays live on the network. Deleting fails only while a post is being published.
How to connect each AI client
Every client uses the same URL. The steps differ by app.
Claude (claude.ai, Desktop and mobile)
- Open Settings → Connectors and click Add custom connector.
- Name it Postika and paste
https://postika.app/api/mcp. - Click Connect, sign in to Postika and click Allow.
Connectors added on claude.ai also appear in Claude Desktop and the mobile apps with the same account. More in the Claude guide.
ChatGPT
- Open Settings → Apps & Connectors → Advanced settings and turn on Developer mode.
- Back in Apps & Connectors, click Create. Paste the URL and choose OAuth authentication.
- Click Create, then Allow in the Postika window. Enable the connector from the composer when you chat.
See the ChatGPT guide.
Claude Code
Run:
claude mcp add --transport http postika https://postika.app/api/mcp
Then type /mcp inside Claude Code, pick postika and authenticate in the browser. Add --scope user to make it available in every project on your machine. Details in the Claude Code guide.
Cursor, VS Code and Gemini CLI
- Cursor: add the server from the app's connection page or in
~/.cursor/mcp.json. Cursor opens the Postika permission window the first time. - VS Code: run MCP: Add Server → HTTP from the command palette and paste the URL, or commit it to your workspace configuration.
- Gemini CLI: add the server, then run
/mcp auth postikainside Gemini CLI.
Agents without OAuth
Other agents connect with a project API key sent as Authorization: Bearer ptk_…. Automation platforms like n8n can also use MCP: an n8n AI Agent with an MCP Client Tool pointed at the same endpoint gets the five Postika tools. The full reference is in the MCP documentation.
Prompts that work well over MCP
Good MCP prompts name the project, the networks, the dates and the mode. Compare:
| Vague | Specific |
|---|---|
| "Post something about our sale." | "In the Acme project, create one Instagram and one Facebook post about the 20% autumn sale, pending approval on Thursday at 12:00." |
| "Do social for next week." | "List next week's posts in Acme. Fill empty weekdays with one LinkedIn post each, rotating our three pillars, pending approval at 9:00." |
| "Write a tweet." | "Write an X post under 280 characters announcing the webinar, and schedule it for 2026-11-12 17:00." |
A useful habit: start sessions with a read-only request ("list what is scheduled") so the AI has context before it writes anything.
A first session, step by step
If you have just connected an AI, this short sequence shows you what it can do without any risk:
- Ask what it can see. "List my Postika projects and their time zones." The AI calls
list_projectsand answers with names and local times. - Check the channels. "Which channels are connected in the Acme project, and which ones need a photo or video?" This uses
list_channels, which returns each channel's character limit and media requirement. - Read the calendar. "What is scheduled for the next 14 days? Only summarize." Read-only, nothing changes.
- Create one test post. "Create a LinkedIn post announcing our webinar next Thursday, pending my approval tomorrow at 10:00." The post appears in the calendar as pending approval.
- Review it in the app. Open the calendar, read the LinkedIn preview, edit if needed and approve, or ask the AI to delete it if it was only a test.
After this, you know exactly how the AI reads your projects and where its posts land. From there, larger requests (a full week, several networks, several brands) work the same way.
Security checklist before you connect an AI
Giving an AI access to your social accounts is a real decision. Check:
- Default mode. Does the server push AI posts to approval, or publish straight away?
- Scope. Which brands or projects can this connection reach? Use API keys per project if you want to restrict an agent to one brand.
- Destructive actions. Is delete marked as destructive so the client asks for confirmation?
- Same rules as the app. Does the server enforce the networks' text and media limits, or can the AI create posts that will fail?
- Revocation. Can you disconnect the AI in one place? In Postika, you can remove any connected AI from the app.
- Who sees what. Read-only tools (list posts) should not leak data across brands that are not in the connection's scope.
Limits of MCP today
- Files: in Postika, media is added in the app, not over MCP.
- Model behavior: the AI can still misread a date or write in the wrong tone. That is why approval matters.
- Client differences: some clients need a developer or advanced setting turned on, and features vary between desktop and mobile apps.
- No analytics through Postika: Postika's MCP tools cover the calendar and publishing; performance data lives in each network's insights.
If you are new to the whole idea, start with our guide to managing social media with AI, then come back to connect a client. Every Postika plan includes the MCP connection and API keys (pricing).
FAQ
What is an MCP server for social media?
It is a service run by a social media tool that exposes actions (list channels, create a post, delete a post) to AI assistants through the Model Context Protocol. Once connected, an AI like Claude or ChatGPT can manage your calendar from a normal chat, following the same rules as the tool's own interface.
Can Claude schedule social media posts?
Yes, when it is connected to a scheduler through MCP. In Claude you add a custom connector with the scheduler's MCP URL, sign in and allow access. Then you can ask it to draft, schedule or list posts, and with Postika the posts wait for your approval by default.
Is it safe to give an AI access to my social media accounts?
It can be, with the right setup. Choose a server where AI posts go to approval by default, where delete asks for confirmation, and where you can limit and revoke access. The AI never gets your social network passwords; it only uses the scheduler's tools you authorized.
Do I need to code to use MCP?
No. In Claude and ChatGPT you paste a URL and sign in. Developers can also use it from Claude Code, Cursor, VS Code or Gemini CLI, and automation builders can call the same server from n8n.