GoodTake AI Docs

Claude.ai Connectors

Connect GoodTake to Claude.ai via the Connectors UI

Setup

  1. Open claude.ai and go to Settings → Connectors
  2. Click Add connector
  3. Enter the MCP URL: https://mcp.goodtake.ai/mcp
  4. Click Connect
  5. A browser popup opens — sign in to GoodTake (or you're signed in already)
  6. Approve the connection — Claude immediately gains access to the GoodTake tools

That's it. No API key to copy.

How the OAuth flow works

Claude.ai acts as an OAuth 2.1 client and follows the standard Authorization Code + PKCE flow:

  1. Claude discovers /.well-known/oauth-authorization-server
  2. Registers dynamically via POST /oauth/register (one-time)
  3. Redirects your browser to /oauth/authorize — you're already logged in via Clerk, so consent is immediate
  4. Exchanges the code at POST /oauth/token for a gt_ API key
  5. Stores the key and uses it on every subsequent /mcp request

The minted key appears in Settings → Developer Settings as Claude MCP – <Month Year>. You can revoke it there at any time to disconnect Claude.

Usage

Ask Claude naturally — it will call GoodTake tools automatically:

"Generate a cinematic landscape, 16:9"

"Create a product image of a white coffee mug on a wooden table"

"Show me my last 5 generations"

Long generations

generate_image waits up to 90 seconds for the image. If it isn't ready by then the tool returns instead of holding the connection open:

"Generation 550e8400-… is still processing. Use get_generation to check its status."

That hand-off is deliberate. Claude.ai cancels a tool call at around two minutes, so returning first turns a slow generation into a clean follow-up rather than a call that looks like it failed. Claude usually calls get_generation on its own; you can also ask:

"Get generation 550e8400-…"

and it will fetch and display the finished images.

Most text-to-image models finish in a few seconds and render inline, so you won't often see this. Editing is the exception — supplying a reference image routes to a slower edit model that takes roughly two to three minutes, so expect the hand-off on almost every edit.