Skip to content

ChatGPT App

Bubbles’ remote MCP server can be added to ChatGPT as a connector, letting ChatGPT read your teams, bubbles, transcripts, and comments — and record a meeting for you — through the tools described in the MCP reference.

Before you start

  • The MCP endpoint is your Bubbles API origin followed by /mcp:

    https://api-tom.usebubbles.com/mcp
  • Adding a connector in ChatGPT currently requires developer mode (Settings → Connectors → Advanced → Developer mode), which lets you add a custom MCP server by URL.

  • Connecting takes no setup: ChatGPT registers itself automatically (dynamic client registration) and completes the OAuth handshake in the background — no token to paste and no approval screen. You sign in afterwards, in plain language, inside the chat.

Add Bubbles in ChatGPT

  1. In ChatGPT, open Settings → Connectors. Enable Developer mode under Advanced if you don’t already see the option to add a custom connector.

  2. Choose Add custom connector (MCP server) and paste your MCP endpoint URL:

    https://api-tom.usebubbles.com/mcp
  3. ChatGPT connects right away — there is no authorization screen to approve.

  4. Start a chat, enable the Bubbles connector, and ask ChatGPT to sign you in (below).

Sign in inside the chat

A newly added connector is not signed in yet: ChatGPT can see the Bubbles tools, but the ones that read your data return a short “not signed in” message with no data until you sign in. There is no password and no token to copy — signing in happens entirely in the conversation:

  1. Ask ChatGPT to sign you in to Bubbles. ChatGPT asks for your email and Bubbles emails you a sign-in code. (The same step logs in existing users and creates an account for new ones.)

  2. Read the code back to ChatGPT. That finishes signing you in — on the same connection.

  3. If you’re new, ChatGPT offers to set up a workspace: join one your email domain already uses, or create your own (which may include a free Pro trial).

Once you’re signed in, the read tools return your data and join_meeting works. ChatGPT remembers the connection, so a new conversation is already signed in.

What ChatGPT can do once connected

The same six read-only tools (full schemas in the MCP reference):

  • list_bubbles, search_bubbles, get_bubble, get_bubble_transcript, get_bubble_comments, list_teams.

plus one write tool:

  • join_meeting — send the Bubbles notetaker to join and record a meeting by link. It uses the meetings:join permission your connection carries and is used only when you explicitly ask ChatGPT to record a meeting.

With the read tools ChatGPT can read your bubbles but cannot create, edit, or delete anything — and it only ever sees bubbles you can read in the product.

Review / availability status

Troubleshooting

  • No “Add custom connector” option — enable Developer mode in Settings → Connectors → Advanced.
  • Discovery / connection fails — confirm you pasted your Bubbles MCP endpoint (https://api-tom.usebubbles.com/mcp) exactly, and that it is reachable over the public internet (ChatGPT connects from OpenAI’s cloud, not your device).
  • Tools return “not signed in” — that’s the expected state before you sign in. Ask ChatGPT to sign you in to Bubbles (above) and read back the emailed code.
  • A tool returns nothing after sign-in — you only see bubbles you can read in the product; the same access rules apply to the API and MCP.