Guides
How to Schedule Social Media Posts With Claude (and Other AI Assistants) Using MCP
· 6 min read
Connect Claude, ChatGPT or Cursor to your social media scheduler with MCP, then draft, queue and schedule posts by chat. Setup, prompts and safety.
AI assistants got good at writing social posts a while ago. What they couldn't do was *post* them. You'd draft a week of content in Claude, then copy each caption into a scheduler, pick channels, pick times, attach images, one at a time. The Model Context Protocol (MCP) removes that last mile. Connect your assistant to your scheduler once, and "queue three posts about our launch for next week" turns into three real, scheduled posts in your calendar.
This guide shows how to connect Claude (and other MCP-capable assistants) to Tesselith, what the assistant can and can't do once connected, prompts that work well, and how to keep it safe when software that writes fluently can also hit publish.
What is MCP, in plain terms?
The Model Context Protocol is an open standard for connecting AI assistants to other software. An app publishes a list of tools ("create a draft", "add to queue", "list channels"), each with a description and a precise input format. When you ask your assistant for something, it decides which tools to call, calls them through the protocol, and reads the results. It's the same idea as a person using an app's buttons, except the buttons are designed for a model to understand.
Because it's a standard, one MCP server works with many assistants: Claude on the web and desktop, Claude Code, ChatGPT connectors, Cursor, VS Code and more. You set it up once per assistant, and the connection is yours to revoke.
Why schedule social posts through an assistant?
- Drafting and scheduling become one step. "Write five LinkedIn posts from this article and queue them" does the writing *and* the slotting.
- Context lives in the conversation. Paste a product brief, a transcript or a changelog, and the assistant turns it into posts that fit each network's limits and tone.
- Planning by question. "What's going out this week? Where are the gaps?" gets an answer from your real queue, not a guess.
- Bulk edits by instruction. "Push everything about the webinar back a week" beats dragging twelve posts around a calendar.
Connect Claude to Tesselith in 2 minutes
Claude on the web or desktop
- Open Settings → Connectors in Claude and choose Add custom connector.
- Name it Tesselith and paste
https://api.tesselith.com/mcp. - Click Connect. Sign in to Tesselith in the window that opens.
- On the consent screen, choose the workspace and review the permissions. Untick anything you don't want; for example, keep *Schedule and queue* off if you only want drafts.
- Click Allow access, then turn the connector on in a chat.
Claude Code
claude mcp add --transport http tesselith https://api.tesselith.com/mcpThen run /mcp in Claude Code to sign in. For a server or CI job where no one can click through a browser, use an API key from Settings → Developers as a bearer header instead. The full MCP setup guide covers other clients.
What the assistant can do once connected
| Ask for… | It uses | Result |
|---|---|---|
| "Which accounts are connected?" | list_channels | Your channels, with any that need reconnecting flagged |
| "What's queued for Instagram?" | get_queue | Upcoming posts and free slots for two weeks |
| "When should I post on LinkedIn?" | suggest_best_time | The next free slot plus your audience's best hours |
| "Draft a post about…" | create_draft | A draft in Tesselith you can open and edit |
| "Add it to the queue" | add_to_queue | The next free slot on every channel it targets |
| "Schedule it for Tuesday 9am New York time" | schedule_post | Pinned to that exact time |
| "Use this image" (with a link) | upload_media_from_url | Uploaded and attached |
| "How are we doing?" | get_analytics_summary | Channels, upcoming, drafts and published counts |
| "Post it right now" | publish_now | Only after you confirm |
| "Delete the duplicate" | delete_post | Only after you confirm |
Every result includes a link straight to the post or queue in Tesselith, so you can check the assistant's work in the calendar with one click.
Prompts that work well
Assistants do best with specific instructions: which channels, how many posts, what voice, and whether to queue or pin times. Some starting points:
- "Here's our launch announcement. Write a version for LinkedIn (professional, 3 short paragraphs), Threads (casual, under 300 characters) and Bluesky (under 300 characters, no hashtags). Create them as drafts. Don't schedule yet."
- "Look at my Instagram queue for the next two weeks. List the gaps, then draft posts from our FAQ page to fill them and add them to the queue."
- "Turn this podcast transcript into a 5-post Threads series and schedule one a day at 9am London time starting Monday."
- "Our webinar moved from Thursday to next Tuesday. Find every scheduled post that mentions it and tell me what you'd change before changing anything."
- "Which channel needs reconnecting, and which scheduled posts will fail because of it?"
Notice the pattern: ask for drafts first, review them in the calendar, then ask the assistant to queue or schedule. It's faster than doing it all by hand and keeps you in the loop on everything public.
Keeping it safe
Letting software publish to your brand's accounts deserves some care. Here's how the Tesselith integration is built to keep you in control, and what to do on your side.
Built-in guardrails
- You approve every permission on the consent screen, grouped into what the assistant can see and what it can change.
- Never more than you. The assistant acts as you, limited to your current role. If your role changes, so does its access.
- Destructive actions need confirmation.
publish_nowanddelete_postare marked destructive and refuse to run without an explicit confirm flag, so a well-behaved client asks you first. - Your social passwords stay private. The assistant never sees network credentials or tokens.
- Everything is logged. Actions taken through MCP appear in the audit log, marked as coming from an AI assistant.
- Instant disconnect. Settings → AI assistants lists every connected assistant with its permissions. Disconnect ends its access on its next request.
Good habits
- Start with drafts only (untick *Schedule and queue*) until you trust the workflow.
- Keep your client's "ask before using tools" setting on for anything that changes data.
- Review assistant-written posts for facts and claims. Models can confidently invent details.
- Give an AI-only workspace or channel a trial run before using it on your main brand account.
- Disconnect assistants you no longer use.
MCP vs. copying from ChatGPT vs. a built-in AI writer
| Copy-paste from a chat | Built-in AI writer | Assistant via MCP | |
|---|---|---|---|
| Writes posts | Yes | Yes | Yes |
| Knows your queue and channels | No | Partly | Yes |
| Schedules and queues | No | No | Yes, with your permission |
| Works across your other tools | No | No | Yes, in the same conversation |
| Effort per post | High | Medium | Low |
Tesselith's built-in AI assistant is still the quickest way to polish a single caption while you're in the composer. MCP shines when the work starts somewhere else, in a doc, a transcript or a chat, and ends with a week of scheduled posts.
Getting started
Add https://api.tesselith.com/mcp as a connector in your assistant, approve the permissions you're comfortable with, and start with a read-only question like "what's going out this week?" From there, have it draft, then queue. The MCP setup guide has instructions for every supported client, and the MCP server page summarises what it can do.
Questions people ask
- Can Claude post to social media?
- Yes, when it's connected to a scheduler through MCP. Connect Tesselith as a custom connector in Claude and it can draft, queue and schedule posts to your connected channels within the permissions you approve.
- Is it safe to let an AI assistant schedule social media posts?
- It can be, with the right guardrails: you approve each permission, the assistant can never exceed your own role, publishing immediately and deleting require confirmation, every action is logged, and you can disconnect it instantly.
- Does this work with ChatGPT?
- Yes. Any assistant that supports remote MCP servers over HTTP can connect, including ChatGPT connectors, Cursor and VS Code. Use the same server URL.
- Do I need to be a developer to use MCP?
- No. In Claude it's a settings screen: add a connector, sign in, approve. Developers can also connect headless tools with an API key.