Skip to main content
Every client needs the same one thing: the MCP URL. The client finds the sign-in page by itself, registers itself, and opens Workchats in your browser. The examples below use Production. For an assistant that can only read, use the URL that ends in /mcp/readonly.
AI clients change their menus often. Each section below says what to paste. Where a menu name is given, it is only a guide. If your client looks different, use its own help for “add a remote MCP server” or “add a custom connector”.

Approve the connection

Every client ends at the same Workchats consent screen:
  1. Sign in to Workchats, if you are not signed in.
  2. Read the assistant’s name and where Workchats sends the sign-in code. For an app on your computer, such as Claude Code, Cursor or VS Code, the screen warns you. Allow it only if you just started the connection from that app.
  3. If you are in more than one company, pick a company. If you can’t connect to a company, the screen shows the reason:
    • your admin hasn’t turned on AI assistants
    • guests can’t connect AI assistants
    • your account is not active
    • your admin has blocked this assistant
    See You cannot connect to a company.
  4. Select Allow. The browser returns you to the assistant.
The connection belongs to that one company. To use a second company, add the server again in the client under another name, for example “Workchats (Corpwise)”, and pick that company.

Claude

Claude on the web and Claude Desktop connect to remote MCP servers as custom connectors.
  1. In Claude, open the connector settings and add a custom connector.
  2. Name it Workchats, and paste https://mcp.workchats.com/mcp as its URL.
  3. Select Connect, then approve in Workchats.
  4. In a chat, turn the Workchats connector on and ask, for example, “Catch me up on my unread messages.”
If an organisation manages your Claude plan, an owner of that organisation may have to add the connector first.

ChatGPT

ChatGPT connects to remote MCP servers as connectors. Which plans can add a custom connector, and whether a workspace admin must allow it first, is decided by OpenAI.
  1. In ChatGPT, open the connector settings and create a connector.
  2. Paste https://mcp.workchats.com/mcp as the server URL, and choose OAuth as the authentication.
  3. Approve in Workchats.
In Deep Research and other modes that use only search and fetch, ChatGPT can read group and channel messages, tasks, meetings and files, but never direct messages. See Search and fetch.

Claude Code

Add the server from a terminal:
Then, in Claude Code, run /mcp, choose workchats and sign in. Claude Code opens Workchats in your browser. A client that receives the sign-in code on your computer, as Claude Code is expected to, makes the consent screen show the local-app warning.

Cursor

Select the button to add the server to Cursor. Cursor asks you to confirm the install. Add Workchats to Cursor For an assistant that can only read, use Add Workchats (read-only) to Cursor. Or add the server to Cursor’s MCP configuration, ~/.cursor/mcp.json for every project or .cursor/mcp.json for one project:
Then open Cursor’s MCP settings and sign in to workchats. If Cursor receives the sign-in code through an app link such as cursor://, the consent screen shows the local-app warning.

VS Code

Select a button to add the server to VS Code. VS Code asks you to confirm the install. Install Workchats in VS Code Install Workchats in VS Code Insiders For an assistant that can only read, use Install Workchats (read-only) in VS Code. Or add the server to .vscode/mcp.json in your workspace, or to your user MCP configuration:
Start the server from VS Code, then sign in when VS Code asks. If VS Code receives the sign-in code on your computer, the consent screen shows the local-app warning.

Codex

The Codex CLI, the Codex IDE extension and the ChatGPT desktop app share one configuration file, ~/.codex/config.toml. Add the server from a terminal:
Or add it to ~/.codex/config.toml yourself, then run codex mcp login workchats:
Codex opens Workchats in your browser and receives the sign-in code on your computer, so the consent screen shows the local-app warning.

Gemini CLI

Add the server from a terminal:
Or add it to ~/.gemini/settings.json yourself:
Then, in Gemini CLI, run /mcp auth workchats. Gemini CLI opens Workchats in your browser and receives the sign-in code on your computer, so the consent screen shows the local-app warning.

The name Workchats shows

Workchats names the assistant on the consent screen, in your company’s admin console, and in the “via” label of the messages it sends for you. Well-known clients get a fixed name: Claude, ChatGPT, Claude Code, Cursor or VS Code. Your admin also sees the client’s block key. A block key that starts with local: belongs to an app on someone’s computer.

Other clients

Any client that supports remote MCP servers over Streamable HTTP with OAuth can connect with the same URL. A client that Workchats does not know shows the name it registered with, cut to 40 characters. Your admin sees that name beside the client’s block key. See OAuth for client developers for what a client must support.

Disconnect

  • From Workchats. Settings → Connected AI apps lists your connections in the current company and disconnects any of them.
  • From your company. A company admin can disconnect any connection.
  • From the client. Removing the server from the client stops the client from calling Workchats. The client may also revoke its token.
A disconnected assistant’s next request fails, and the client asks you to connect again.

Connections that end by themselves

You must connect again when:
  • the assistant was not used for 30 days
  • your company’s maximum session lifetime passed since you connected
  • your company turned AI assistants off, or blocked the assistant
  • your account stopped being an active employee
  • you connected the same client again through the same sign-in route. The new connection replaces the old one.
Turning AI assistants on again, or unblocking a client, does not bring back an ended connection.

Something went wrong

See Troubleshooting for common problems, and for how to contact support.