Browse documentation
All documentation

AI assistants (MCP)

Manage connected assistants and tokens

Review and disconnect connected assistants, connect coding agents with a setup prompt and personal access token, and fix common connection problems.

Last reviewed October 15, 2026

Everything about your assistant connections lives in one place: open your account menu, choose Settings, then select MCP.

What you see in Settings > MCP

  • MCP endpoint is the server URL to paste into your assistant.
  • Connect a client shows sign-in steps for Claude and Claude Code, and a ready-made setup prompt for other coding agents.
  • Personal access tokens create and list the tokens that setup prompts use.
  • Connected apps lists every assistant you approved through Fuzzbucket sign-in.

Disconnect an assistant

In Connected apps, each entry shows when it was connected, when it was last used, and which brands it can reach. Select Disconnect and confirm. The assistant loses access immediately and must sign in again to use Fuzzbucket.

You can also remove the connector inside your assistant. Disconnecting in Fuzzbucket is the reliable way to make sure access has ended.

Connect a coding agent with a setup prompt

Claude, ChatGPT, and Grok connect by signing in, so they never need a token. For coding agents such as Codex, Cursor, VS Code, Gemini CLI, or Grok Build, a setup prompt is the quickest route: you paste one message and the agent configures itself.

  1. In Connect a client, choose your agent.
  2. Under Token to include, pick an existing token. To make a new one, go to Personal access tokens, choose the Permissions (Read content, Create and edit, or Full access), a name, and an expiry, then select Create token.
  3. Select Copy setup prompt. When you create a new token, the prompt is usually copied for you.
  4. Paste the prompt into your agent. It adds the Fuzzbucket server with the token, stores the token privately, and checks that the connection works.

The prompt contains the server URL and the token, so a developer can also use them to configure a tool by hand with an Authorization: Bearer header.

A token works like a password. Paste the setup prompt only into your own agent, never into chats, emails, or support requests. If a token might be exposed, select Revoke next to it in the token list and create a new one.

Troubleshooting

  • Sign-in does not finish. Check that the server URL is exactly right, that pop-ups are allowed, and that you are signed in to the Fuzzbucket account you want. If it still fails, remove the connector and add it again.
  • Permission denied or insufficient scope. The connection was approved with a lower access level than the request needs. Disconnect, connect again, and choose a higher level, or create a token with the right permission.
  • Token rejected. Tokens stop working when they expire or are revoked. Create a new one.
  • Out of credits. Chat and generation stop when you have no credits left, but reading still works. Add credits or change your plan, then try again. See Credits and generation cost.
  • Request refused. Fuzzbucket or a model provider declined the request under its content rules. Rephrase the request or change the reference material.

Still stuck? Email hi@fuzzbucket.ai with the assistant you use and roughly when the problem happened. See Report a problem for what else to include.