Connect Any AI Assistant to ArchySocial: The Complete MCP Setup & Usage Guide
Wire ChatGPT, Claude, VS Code, and Gemini straight into ArchySocial with one OAuth sign-in — then let them generate, upload, and draft real posts for your connected networks.
AI6 min read
The best social tool is the one you never have to open. Connect your AI assistant once, and “post this for me” becomes a sentence — not a workflow.
ArchySocial already turns one brief into campaign-ready posts. But most of your ideas don’t start in ArchySocial — they start in a conversation with ChatGPT, a thread in Claude, a coding session in VS Code, or a quick question to Gemini. The MCP connector closes that gap: any MCP-capable assistant can create campaigns, generate images and video, and prepare posts for your connected networks through natural conversation — and everything lands safely as a draft in your account.
This guide covers both halves of the story: how to connect your assistant in a single OAuth sign-in, and how to use it so posts come out exactly the way you expect.
What “connecting an assistant” actually means
The Model Context Protocol (MCP) is a standard way for an AI assistant to call tools on your behalf. ArchySocial runs a remote MCP server at a single endpoint, https://archy-social.archy-ai.com/mcp, that exposes 38 tools — everything from generating captions to publishing a post. Once you connect, your assistant can call those tools in plain language, and every call runs inside your account.
Three things happen under the hood: your assistant authenticates as you via OAuth, it calls the /mcp endpoint with the token it was issued, and the tools act only within your workspace and your connected networks. No scraping, no screen control — just scoped API calls.
What your assistant can do
Generate
Per-network captions and hashtags, on-brand images, and short vertical video — produced in the background with Azure OpenAI.
Upload & attach
Push an image or video you already have straight into your library and attach it to a post — no copy-pasting URLs.
Draft, schedule, publish
Prepare posts as drafts, schedule them for later, or publish immediately to LinkedIn and X.
Read & analyze
List campaigns, posts, and engagement metrics so the assistant can reason about what’s actually working.
One sign-in: how OAuth 2.1 works here
You never paste an API key into ChatGPT. The first time your assistant uses a tool, it runs a standard OAuth 2.1 sign-in against your ArchySocial login — the same pattern your bank’s “connect an app” flow uses.
Why go through a browser sign-in instead of a key you paste once? Because it is dramatically safer and simpler:
No shared secrets — the assistant receives a scoped, expiring token, never your password or a long-lived key.
Self-registration — clients register themselves via Dynamic Client Registration, so there is no client ID to copy by hand.
PKCE S256 — the flow is hardened against authorization-code interception.
Revoke anytime — disconnect from ArchySocial → Account → Connected apps and the token stops working instantly.
Connect your client
Every client below talks to the same endpoint: https://archy-social.archy-ai.com/mcp. Pick yours.
ChatGPT
Open Settings → Connectors (or Developer mode → Add custom MCP server).
Paste the server URL: https://archy-social.archy-ai.com/mcp
Leave authentication as OAuth — ChatGPT requires it and does not accept custom API keys.
Use a tool once; the ArchySocial sign-in and consent screen opens. Approve it, and you’re connected.
Claude
Desktop or mobile: Settings → Connectors → Add custom connector.
Enter the URL https://archy-social.archy-ai.com/mcp and sign in when prompted.
Claude Code (CLI): run claude mcp add --transport http archy-social https://archy-social.archy-ai.com/mcp then type /mcp and choose Authenticate.
VS Code (GitHub Copilot)
Add a server entry to .vscode/mcp.json (or your user settings) — just the URL, no token:
"servers": { "archy-social": { "type": "http", "url": "https://archy-social.archy-ai.com/mcp" } }
Start it from MCP: List Servers → archy-social → Start. VS Code discovers the authorization server from the endpoint’s challenge and opens the browser sign-in for you.
Gemini CLI
Add the server to ~/.gemini/settings.json:
"mcpServers": { "archy-social": { "httpUrl": "https://archy-social.archy-ai.com/mcp" } }
Then run /mcp auth archy-social to trigger the browser sign-in — Gemini CLI discovers the authorization server automatically.
Prefer an API key?
Headless tools, CI jobs, and scripts can skip the browser. Generate a key in ArchySocial → Account & Connections → MCP API Keys, then send it as an Authorization: Bearer archy_sk_… header. (ChatGPT is the exception — it only supports OAuth.)
How to use it: the draft-first pipeline
Here is the part that saves you from surprises. ArchySocial is deliberately conservative: anything your assistant creates lands as a draft, not a live post. Nothing publishes on its own.
When you ask for a post with a visual, the assistant generates (or uploads) the media, gets back a URL, and attaches that URL to the post. The post is saved as a draft — the publisher will not touch it until you decide. From there, you have three moves:
Schedule it — give it a time and it joins the publish queue for its network.
Publish it now — the assistant can push it live immediately on request.
Open it in ArchySocial Studio — tweak the caption, swap the image, then schedule.
And posts are only ever prepared for networks you have actually connected. Ask for a TikTok post without a TikTok connection, and the assistant is told to connect it first — no silent dead-ends, no orphaned posts.
Prompts that work well
Draft a launch
“Draft a LinkedIn and X post about our new pricing, generate an on-brand image, and save them as drafts.”
Reuse your media
“Upload this screenshot and attach it to my latest LinkedIn draft.”
Ship it
“Publish the approved X draft now.”
Close the loop
“Which of last week’s posts got the most engagement? Draft three follow-ups in that style.”
Permissions, tenancy, and safety
The connector is scoped on three axes, so an assistant can never do more than you intend.
Scopes
Consent grants mcp:read, mcp:write, and profile. A read-only connection can browse campaigns and analytics but never publish.
Your role is the ceiling
Effective access is the lesser of the granted scope and your team role. A viewer can’t publish even if mcp:write was granted.
One tenant
A connection is bound to a single workspace, so cross-account access is impossible by construction.
Tokens are hashed
Access and refresh tokens are stored only as SHA-256 hashes and are never written to logs.
When something doesn’t connect
“Can’t find the server”
Make sure you used the /mcp URL — not the bare domain. Discovery only works against the endpoint.
Stuck on sign-in
You need an active ArchySocial account and team membership. Sign in to the app in a browser first, then retry.
Write tools denied
Your connection is read-only, or your team role is viewer. Reconnect with mcp:write and a publishing role.
Won’t post to X / LinkedIn
The underlying social account isn’t connected in ArchySocial, or the workspace is out of credits.
Give your assistant social superpowers
Connect ChatGPT, Claude, VS Code, or Gemini in one sign-in — then just ask. Full AI generation is free for everyone until June 1, 2026.
