David C.

(We've improved our docs in this area https://motherduck.com/docs/key-tasks/ai-and-motherduck/mcp-workflows/connect-with-access-token/, let us know if you have any feedback!)

Perfe, let us know how i goes! We're doing a cleanup of our docs regarding this 🙂

Oki. Yeah it's a bit annoying the way to get this to work at the moment. We are tracking improvements to support this use case, I've mentioned your case in our internal tracker 😊 . The way to get this to work it to configure the agent client to authenticate via a bearer token, and passing a SA token to that. If you use claude code or codex you can auth the MCP server with your SA token directly:

export MOTHERDUCK_TOKEN='<your-SA-token>'

codex mcp add motherduck-sa \
  --url https://api.motherduck.com/mcp \
  --bearer-token-env-var MOTHERDUCK_TOKEN

claude mcp add --transport http --scope user \
  motherduck-sa https://api.motherduck.com/mcp \
  --header 'Authorization: Bearer ${MOTHERDUCK_TOKEN}'

For Claude GUIs (desktop, web) you can also add a custom connector if your account supports configuring request headers. 1. On Customize > Connectors > Add custom connector. 2. Enter https://api.motherduck.com/mcp. 3. Choose No sign-in. 4. Under Request headers, add: ◦ Name: Authorization ◦ Value: Bearer <your-SA-token> 5. Save. If your account doesn't support the headers yet, for Claude Desktop only it's also possible to add a local redirect by adding the following to your claude_desktop_config.json.

"mcpServers": {
    "MotherDuck (prod)": {
      "command": "npx",
      "args": [
        "mcp-remote",
        "https://api.motherduck.com/mcp",
        "--header",
        "Authorization: Bearer <your-SA-token>"
      ]
    }
  },

Finally (😅), for ChatGPT you need to create a custom MCP app (docs), and choose "Access token /API key" auth method. I think that option might only be available for some accounts though...

What are you trying to do? depending on the use case, using the MotherDuck CLI might easier and more flexible which agent client are you using? --- There's a straight forward way if your client supports bearer tokens, you can pass the SA token in the header:

{
  "headers": {
    "Authorization": "Bearer <your_SA_token>"
  }
}

I don't think there's tight competition at the moment 😅

knowyourmeme.com/memes/friendship-ended-with-mudasir

Have a nice weekend!

Ok! I'll share some high level guidance, documentation references and answer your direct questions. I think your team has a call already scheduled with our team for next week to go over in detail about how to accomplish what you're looking to do (?) In any case, very happy to answer any follow-up questions right now 🙂 The first step would be to get the data into motherduck. "Millions of rows" is not that much volume nowadays, depends on how many millions and the size of each record.

Then you could optionally transform the raw data from postgres into other tables, or build your reports directly from the postgres-shaped data. To create the reports, it seems like you're looking to embed dives into your own application? The general docs about that are here: motherduck.com/docs/key-tasks/dives and motherduck.com/docs/key-tasks/dives/embedding-dives For your users editing Dives:

Can I integrate MotherDuck in my application so that user don't have to go to Claud and can create Dives from my app ?

Dives under the hood are front-end JS code with functions that can call MD to get the data to visualize. So they can be created by manually writing the code, using agents, or by having static or parametrized Dives that are generated by your application. The functions that are available to interact with Dive content programatically are documented here: motherduck.com/docs/sql-reference/motherduck-sql-reference/dives