How to let AI run your social media: Claude, Codex, Cursor & more (Full MCP Setup)

Your AI assistant can already write your posts.

It just can't post them. So you copy, paste, open 6 tabs, upload the same image 6 times, and fix the character count for X by hand.

We built Autoadify's MCP server to kill that copy-paste step. It went live this week.

Here's what it gets you:

• Claude, Codex, Cursor, Gemini CLI or VS Code posts to Instagram, Facebook, LinkedIn, X, TikTok, YouTube, Reddit, Bluesky and Threads
• Different text per platform, from one request
• Your media library, searchable by what's actually in each image
• Everything lands in your Autoadify calendar, so you can still see and edit it

Let's get into it.

1. Get a key (2 minutes)

In Autoadify, go to Settings → API & MCP and click Create key.

Give it Read and Publish.

Pro tip: limit the key to the channels you want the assistant to touch. Start with one test channel. Publishing is real, so it goes out on your actual accounts.

The key starts with aak_live_. You only see it once. Put it in your password manager.

(API & MCP are on paid plans, from Pro at $29/mo.)

2. Connect your assistant

One server URL for everything:

https://mcp.autoadify.com/mcp

Claude Code (one line):

claude mcp add --transport http autoadify https://mcp.autoadify.com/mcp --header "Authorization: Bearer aak_live_..."

Cursor: add it to ~/.cursor/mcp.json:

{
  "mcpServers": {
    "autoadify": {
      "url": "https://mcp.autoadify.com/mcp",
      "headers": { "Authorization": "Bearer aak_live_..." }
    }
  }
}

Codex (OpenAI's coding agent). Put your key in an env var, then:

export AUTOADIFY_API_KEY="aak_live_..."
codex mcp add autoadify --url https://mcp.autoadify.com/mcp --bearer-token-env-var AUTOADIFY_API_KEY

Gemini CLI: add it to ~/.gemini/settings.json:

{
  "mcpServers": {
    "autoadify": {
      "httpUrl": "https://mcp.autoadify.com/mcp",
      "headers": { "Authorization": "Bearer aak_live_..." }
    }
  }
}

VS Code, Claude Desktop and any other client that supports remote MCP with a custom header work too. The exact config for each is in the docs (link at the end).

(Using the ChatGPT app? It can't send an API key header to a custom server yet, so use Codex for now.) [CHECK BEFORE POSTING]

3. Ask it what it can see

Before you post anything, run this:

Prompt:

What channels do I have connected, and what does each one need before I can post to it?

This calls list_channels. You get every account plus its rules: Instagram needs an image or video, X caps your characters, Reddit needs a subreddit.

Now your assistant knows the rules before it writes a single word.

That alone fixes half the "why did this fail" moments.

4. Write once, post everywhere (properly)

This is where it clicks.

Prompt:

Here's our announcement: [paste it].
Schedule it to LinkedIn and X for Tuesday 9am IST.
Keep the full version for LinkedIn. Write a punchier version under 280 characters for X.
Show me both versions and the time before you schedule anything.

One post, two versions, one request. Under the hood it's a single post with per-platform text, so it shows up as one item in your calendar.

The "show me before you schedule" line matters. Good assistants confirm first anyway, but say it out loud.

5. Let it find the image for you

Most media libraries are a pile of IMG_4471.JPG.

Ours describes what's actually in each file, so the assistant can search by content.

Prompt:

Find a photo from our library that shows the product on a desk, and use it for an Instagram post tomorrow at 6pm about our new feature.

Don't have the image in the library yet? Give it a link:

Import this image and post it to Instagram tomorrow morning: [URL]

If you're in Claude Code, it can even upload a file straight from your computer.

6. Run your week from one chat

Once it's connected, the calendar is a conversation.

Prompts I use every Monday:

What's scheduled for next week?

Move Friday's post to Monday, same time.

Did yesterday's posts go out? If any failed, tell me why.

That last one is underrated. Every post comes back with its status per platform, the live link, and the exact error if something failed (like a connection that needs reconnecting).

And the errors are written so the assistant can actually fix the request, not just say "400".

7. Guardrails (read this)

It's your real accounts, so:

• Scope the key. A read-only key gives the assistant zero tools that change anything. Great for a reporting agent.
• Limit channels. Give a test key one channel until you trust the flow.
• Same rules as the app. Platform requirements, plan limits and our content policy all still apply.

That's it

No more copy-paste between 6 tabs.

You write (or the AI writes), and the AI schedules it.

Docs + every tool: https://docs.autoadify.com/mcp/overview