Docs · Agent API
Connect your assistant to CreatorRM
Use the AI assistant you already have: Claude, ChatGPT, Gemini, Grok, Copilot, or a private model on your own computer. It connects over the Model Context Protocol (MCP), reads your workspace data, and acts inside CreatorRM with your own role, never more.
Paste this address into your assistant:
https://mcp.creatorrm.app/mcp
There is no key to copy. Your assistant opens a CreatorRM sign-in the first time it connects.
Server details
- MCP endpoint
https://mcp.creatorrm.app/mcp- Server
https://mcp.creatorrm.app. The bare server address also accepts MCP requests, so either works.- Transport
- Streamable HTTP.
- Sign-in
- OAuth 2.1 with PKCE (S256). Assistants register themselves (dynamic client registration), so there is no client ID or secret to set up.
- Discovery
https://mcp.creatorrm.app/.well-known/oauth-protected-resourcehttps://mcp.creatorrm.app/.well-known/oauth-authorization-serverhttps://mcp.creatorrm.app/.well-known/mcp/server-card.json
How connecting works
- Add the endpoint in your assistant, using the steps for it below.
- Your assistant opens a CreatorRM sign-in window. If you are already signed in, you go straight to the permission screen.
- Pick the workspace, review the permissions the assistant asks for, and approve. CreatorRM shows this screen for every new connection and never grants access silently.
- The assistant gets access to that one workspace. Access renews itself while you use it. A connection left unused for 30 days has to sign in again.
To disconnect, open Settings → Integrations → Connect AI Assistants in CreatorRM and choose Disconnect under Connected apps. Owners and admins can disconnect any member's app.
What your assistant can do
- Every connection belongs to the person who signed in, and to one workspace.
- The permission screen only offers what your role allows. Admin-level permissions can only be granted by an owner or admin.
- CreatorRM checks permissions on its own servers for every tool call. Your assistant acts with your own role: an editor's assistant can do what an editor can do, and nothing more.
- The workspace owner's assistant works as soon as it connects. Today, another member's assistant needs the owner or an admin to allow it on Settings → AI → AI Delegation, which can also narrow what it may do.
Actions that can't be undone
A connected assistant can read content that came from outside your workspace, such as inbound email, so CreatorRM does not leave its riskiest actions to the assistant's judgment.
- Sending email, SMS, WhatsApp, Slack or Telegram messages, inviting guests to live sessions, and running automations are off for connected assistants until a workspace admin turns them on under Settings → Integrations → Connect AI Assistants → Tool permissions.
- Permanent deletes are off for connected assistants unless a workspace admin allows them.
- Where deleting is allowed, deleting a transcript takes two steps. The first returns a preview of what will be destroyed and a single-use confirmation code that expires after two minutes. Your assistant has to show you the preview and come back with the code once you confirm.
Set up your assistant
Every guide uses the same address, https://mcp.creatorrm.app/mcp. If a menu name below doesn't match what you see, look for the assistant's connectors or MCP settings; these apps rename menus often.
Claude (web and desktop app)
- Open Customize → Connectors, click +, then Add custom connector.
- Name it CreatorRM, paste the address, and click Add.
- Click Connect and sign in. Turn it on in a chat from + → Connectors.
On Team and Enterprise plans an owner adds it first under Organization settings → Connectors → Add → Custom → Web; each member then clicks Connect. Without a paid Claude plan you can add one custom connector.
Claude Code
Run this in your terminal:
claude mcp add --transport http --scope user creatorrm https://mcp.creatorrm.app/mcpThen run /mcp inside Claude Code, pick creatorrm, and sign in. --scope user makes it available in every project; leave it out for the current project only, or use --scope project to share the setup with your team through .mcp.json (each person still signs in as themselves).
ChatGPT (Developer mode)
CreatorRM can change data as well as read it, so ChatGPT needs Developer mode, which is on paid plans. On a Business or Enterprise workspace an admin may need to allow it first.
- In Settings, open Apps & Connectors → Advanced settings and turn on Developer mode.
- Create a connector: name it CreatorRM, paste the address, choose OAuth, and save. Sign in.
- In a chat, click + → More → Developer mode and pick CreatorRM. ChatGPT turns connectors on per chat.
Gemini app and Gemini CLI
Gemini app. Custom apps need a personal Google Account, age 18 or over, the US, and Keep Activity turned on.
- On gemini.google.com, open Settings → Connected apps.
- Under Custom apps, choose Add a custom app and paste the address.
- Sign in and approve.
Gemini CLI. Run:
gemini mcp add --transport http --scope user creatorrm https://mcp.creatorrm.app/mcpThen run /mcp auth creatorrm in Gemini CLI to sign in.
Grok
- Go to grok.com/connectors, click New Connector, and choose Custom.
- Paste the address and complete the sign-in Grok opens.
Custom connectors depend on your Grok plan.
Microsoft Copilot Studio
- Open your agent's Tools page, choose Add a tool → New tool → Model Context Protocol.
- Enter a server name (CreatorRM), a short description of what it holds (contacts, content, calendar, campaigns), and the address.
- Set authentication to OAuth 2.0 with type Dynamic discovery, then Create and Next.
- Create a new connection, sign in, and choose Add to agent.
Local models: Goose, LM Studio, Jan, Open WebUI with Ollama
Your model and its inference stay on your computer. Only the tool calls your assistant makes reach CreatorRM, and CreatorRM never connects to your computer. Choose a model that calls tools reliably; small models often don't.
Goose.
- Run
goose configure, choose Add Extension, then Remote Extension (Streamable HTTP). - Name it creatorrm and paste the address. Goose Desktop shares this setup.
- Start a session; Goose opens the CreatorRM sign-in the first time it uses the extension.
LM Studio.
- In the right sidebar, open the Program tab, then Install → Edit mcp.json.
- Add CreatorRM and save:
{
"mcpServers": {
"creatorrm": {
"url": "https://mcp.creatorrm.app/mcp"
}
}
}Approve the sign-in LM Studio opens in your browser. Signing in needs a recent LM Studio version.
Jan.
- Open Settings → MCP Servers and click + Add MCP Server.
- Choose the HTTP transport, paste the address, and switch the server on.
- Click Sign in on the server's Auth line and approve in your browser.
Open WebUI with Ollama models.
- As an admin, open Settings → Admin → Integrations and click + Add Connection under External Tool Servers.
- Set Type to MCP (Streamable HTTP), paste the address, set Auth to OAuth 2.1, click Register Client, then Save.
- In a chat, open + → Integrations → Tools, turn on CreatorRM, and sign in. Each person signs in as themselves.
Open WebUI needs version 0.6.31 or later and a fixed WEBUI_SECRET_KEY, or sign-ins are lost when it restarts. Reach Open WebUI over https:// or on localhost: CreatorRM refuses sign-in callbacks over plain http to any other address.
Cursor and other MCP clients
Any client that supports Streamable HTTP and OAuth can connect. For clients configured with an mcp.json file, such as Cursor:
{
"mcpServers": {
"creatorrm": {
"url": "https://mcp.creatorrm.app/mcp"
}
}
}Troubleshooting
- "No scopes requested" when you sign in. Your assistant didn't ask for any permissions. Check its connector settings for a scope field.Rolling outCreatorRM will offer a default set of permissions when an assistant asks for none.
- A member's assistant connects but every tool is refused. Today a member other than the owner needs an AI Delegation from the owner or an admin (see What your assistant can do).
- Sending or running an automation is refused. Those actions are off for assistants until an admin turns them on under Tool permissions.
- A self-hosted client can't finish signing in. Its callback address must use https:// or localhost.
For developers
Call the server directly with an access token:
POST https://mcp.creatorrm.app/mcp
Authorization: Bearer <access_token>
Accept: application/json, text/event-stream- Send
initialize, thentools/list. The tool list is live and authoritative; the count on the server card is approximate. - A
401carriesWWW-Authenticate: Bearer resource_metadata="…", which points at the protected-resource metadata so a client can find the authorization server on its own. - Access tokens last one hour. Refresh with
grant_type=refresh_token; refresh tokens rotate on every use and expire after 30 days. Reusing an old refresh token revokes the whole connection. - Honor the retry hints on
429responses.
More for agents: the API catalog, the connect-and-produce skill, and llms.txt.