Skip to main content
Connect an AI agent to Sydium and run your accounts by talking to it - draft in your brand voice, publish, schedule, cancel, and pull analytics. The connect is one browser click: no API keys to paste, no developer app to register. Sydium’s MCP server is a real OAuth 2.1 server with dynamic client registration, so the agent registers itself and you just approve the scopes. Works with Claude, ChatGPT, Cursor, or any MCP client. Endpoint

​
How it works

Your agent gets 9 tools. Connecting takes three steps - nothing to install:
1

Authorize once

Point your client at the Sydium MCP endpoint and click Authorize in the browser. No developer app to register, no API keys to paste - Sydium registers your client automatically, and you approve exactly which scopes it gets.
2

Your agent discovers the tools

On connect, the agent reads the 9 Sydium tools and their schemas - so it knows precisely what it can do and how to call each one.
3

Command in plain language

Ask for what you want - “publish this everywhere,” “schedule the recap for Friday.” The agent calls the right tools and reports back. You keep the final say, always within the scopes you approved.

​
Available tools

Agencies: call list_clients to get your clients, then pass a client’s id as the clientId argument to publish_post, list_accounts, or list_scheduled_posts to act on that client’s connected accounts instead of your team’s own.

Tools reference

Full input schema and behavior for every tool.

​
Authentication

Your client connects over OAuth: it sends you to a Sydium consent screen, you pick Authorize, and the agent gets a scoped token. The agent never sees a password, and it can never do more than the scopes you grant - or more than your own team role allows. The Client setup guide has the per-client steps and a walkthrough of the consent screen.

​
Connect your client

Client setup

Step-by-step for Claude Desktop, ChatGPT, and Cursor.

​
Quick example

Once connected, just talk to your agent:
The agent calls list_accounts, generate_caption, and publish_post (with scheduleAt), then reports back. You keep the final say.

​
FAQ

No. MCP uses one-click browser OAuth with dynamic client registration. API keys are only for the REST API and CLI.
Any MCP-compatible client - Claude (desktop and web), ChatGPT, Cursor, and others. See Client setup.
Yes. add_media stores an image or video and returns the URL that publish_post needs, and it tells you which platforms will accept the file before you schedule anything. Give it a public link to the file, or - in ChatGPT on the web - drag the file straight into the conversation. Claude connectors cannot pass attachments yet, so there a link is the way in.
Only the scopes you approve on the consent screen, and never more than your own team role allows.