Octonity speaks MCP: connect Claude to your social media workspace

For the past two years, "AI in social media tools" has mostly meant one thing: the tool grew a text box that writes captions. Useful, but it quietly assumes the assistant lives inside the tool.

That assumption is flipping. Most people who work with an AI assistant daily already have one they trust — Claude on the desktop, Claude Code in the terminal, Cursor in the editor — loaded with their context, their style, their running projects. The interesting question is no longer "does your social tool have AI?" It's "can your assistant operate your social tool?"

As of today, with Octonity, it can. Octonity now exposes an MCP server — in beta, on Business and Agency plans — so the assistant you already use can read your real workspace and act in it: list your channels, pull analytics, scan the inbox, schedule and publish posts, reply to messages. You bring the assistant; Octonity becomes something it can operate.

What is MCP, in 60 seconds

MCP — the Model Context Protocol — is an open standard that lets an AI assistant connect to an external tool and use it: read its data, call its actions. Think of it as a USB port for assistants. The tool (here: Octonity) runs an MCP server that describes what it can do; the assistant (Claude, Cursor, your own agent) is the client that discovers those capabilities and calls them when your request needs them.

The practical upshot: you connect once, with a token you control, and from then on the assistant doesn't guess about your workspace — it looks. "What's scheduled this week?" becomes an actual query against your actual calendar, not a hallucinated answer.

What you can actually ask

The best way to explain it is to show it. These are the kinds of exchanges the connection makes possible — each backed by a real tool call, not by the model's imagination.

"What's going out this week?" The assistant calls list_scheduled with a Monday-to-Sunday window and gets back your scheduled posts, drafts, and recently published items. Follow up with "anything missing for the product launch?" and it reasons over the real list.

"Anything in the inbox need a reply?" get_inbox returns DMs, comments, and mentions across your connected channels. The assistant can triage — "two DMs look like support questions, one comment is a complaint about shipping" — and draft replies for you to approve. When you say "send that one," it calls reply_to_message to answer in the actual thread.

"How did last month compare to the month before?" get_analytics pulls impressions, engagement over time, per-channel comparisons, top posts, and follower counts. Ask for the comparison in plain language, or ask the assistant to turn it into a client-ready summary — it's working from your numbers either way.

"Schedule this for Thursday at 9." You paste or co-write the copy in your assistant, then it calls schedule_post to queue it — or publish_post if you want it live now. One deliberate guardrail: the server instructs the assistant to confirm the exact wording with you before anything is published publicly. The assistant drafts; you say yes.

The seven tools

The beta ships seven tools — four that read, three that write:

| Tool | Type | What it does | | --- | --- | --- | | list_channels | Read | Your connected accounts and brands | | list_scheduled | Read | The calendar in a date window — scheduled, drafts, published | | get_analytics | Read | Impressions, engagement over time, per-channel comparison, top posts, follower counts | | get_inbox | Read | DMs, comments, and mentions across channels | | publish_post | Write | Publish a post immediately | | schedule_post | Write | Queue a post for a future slot | | reply_to_message | Write | Reply in a thread — DM or public comment |

That's deliberately small. We'd rather ship seven tools that work than thirty that half-work, and expand from real usage.

How to connect

First, create a token in Octonity: Settings → Access tokens → New token. Tick only the scopes you need; the token is shown once, so copy it then. Tokens look like oct_pat_….

Claude Code or the VS Code extension

One command:

claude mcp add --transport http octonity https://api.octonity.com/mcp \
  --header "Authorization: Bearer oct_pat_…" --scope user

Claude Desktop or claude.ai

Go to Customize → Connectors → Add custom connector, set the URL to https://api.octonity.com/mcp, choose fixed credentials as the authentication method, and add the header Authorization: Bearer oct_pat_….

Cursor and the OpenAI API (via the Responses API, for developers building their own agents) connect the same way: HTTP transport, bearer token in the Authorization header.

Then just ask something: "List my Octonity channels." If the assistant comes back with your actual accounts, you're connected.

Security: what the token can and can't do

Handing an AI assistant keys to your social accounts deserves a real answer, not a reassuring sentence. Here's the model:

• Five granular scopes. analytics:read, inbox:read, posts:read, inbox:write, posts:write. A token only carries the scopes you ticked. If you want an assistant that can analyze but never post, grant the read scopes and stop there.
• One token, one workspace. Each token is bound to one user in one workspace and cannot switch workspaces. For agencies this matters: a token created in client A's workspace structurally cannot touch client B.
• Your role is the ceiling. The token acts with your permissions, never more. If your role can't publish in Octonity's approval workflow, neither can any assistant using your token.
• Shown once, stored hashed. Like GitHub and Stripe tokens, we store only a hash — nobody, including us, can read your token back out of the database.
• Instantly revocable. Delete it in Settings → Access tokens and it stops working immediately.
• Plan checked on every call. Not at token creation — on every single request. A downgrade from Business cuts off access at once.

What's not there yet

Beta means beta, so here's the honest list:

• ChatGPT doesn't work yet. The ChatGPT app only accepts OAuth 2.1 connectors, and Octonity's MCP server uses pasted bearer tokens today. ChatGPT support is in development. What works now: Claude Code (CLI and VS Code extension), Claude Desktop, claude.ai on the web, Cursor, and the OpenAI API for developers.
• Business and Agency plans only. Free and Creator plans can't create an access token. See pricing for what's on each plan.
• Seven tools, not everything. You can't yet manage team members, edit an existing scheduled post in place, or run the approval flow itself through MCP. The tool set will grow; the current seven cover the daily loop of planning, publishing, inbox, and reporting.

If something misbehaves, tell us — that's what the beta label is for.

FAQ

What is an MCP server for social media?

An MCP server is a service that exposes a tool's data and actions to AI assistants over the open Model Context Protocol. A social media MCP server, like Octonity's, lets an assistant such as Claude read your content calendar, analytics, and inbox, and schedule or publish posts — against your real accounts, with a token you control.

Does the Octonity MCP server work with ChatGPT?

Not yet. The ChatGPT app only accepts connectors that authenticate with OAuth 2.1, while Octonity currently uses pasted bearer tokens. ChatGPT support is in development. Today it works with Claude Code, the Claude Code VS Code extension, Claude Desktop, claude.ai, Cursor, and the OpenAI API (Responses API) for developers.

Which Octonity plans include MCP access?

Business and Agency plans. Free and Creator plans cannot create an access token, which is required to connect. The feature is in beta on those plans — see the pricing page for details.

Is it safe to let an AI assistant post to my accounts?

You decide how much it can do. Tokens carry only the scopes you grant — read-only is a perfectly good setup — and a token never exceeds your own role's permissions. The assistant is instructed to confirm exact wording with you before anything publishes publicly, tokens are stored hashed and revocable instantly, and your plan is re-checked on every call.

How do I connect Claude to Octonity?

Create a token under Settings → Access tokens in Octonity, then either run claude mcp add with the HTTP transport and your bearer token (Claude Code), or add a custom connector in Claude Desktop or claude.ai pointing at https://api.octonity.com/mcp with fixed credentials. The full commands are in the "How to connect" section above.

The bottom line

The shift isn't that Octonity got a chatbot — it already has AI features where they belong. The shift is that the assistant you already work with all day can now operate Octonity directly: check the week, triage the inbox, pull the numbers, queue the post. If you're on a Business or Agency plan, create a token, run one command, and ask your assistant what's going out this week. It'll tell you — from your actual calendar.