Skip to content
SparkleTree SparkleTree Docs

Connect SparkleTree to your AI

You can give your AI assistant a direct line to your SparkleTree workspace. Once connected, you ask in plain language (“how did my summer campaign do last week?”) and your AI does the clicking for you.

The connection uses MCP, the standard protocol for connecting assistants to tools. Every client below uses the same address:

https://app.sparkletree.io/api/mcp

Before you start

Section titled “Before you start”

You need a SparkleTree account. An AI can’t sign you up: create your account and workspace at app.sparkletree.io first, then come back here.

Connect Claude (Claude.ai or Claude Desktop)

Section titled “Connect Claude (Claude.ai or Claude Desktop)”

SparkleTree isn’t listed in Anthropic’s connector directory yet, so today you add it as a custom connector. It takes about a minute:

  1. In Claude, open Settings → Connectors.
  2. Click Add custom connector.
  3. Paste this URL: https://app.sparkletree.io/api/mcp
  4. Approve the connection when Claude asks.
  5. A browser window opens. Sign in with your SparkleTree account and confirm.

That’s it. Start a new conversation and ask Claude something about your workspace to check it’s working.

Connect ChatGPT

Section titled “Connect ChatGPT”

ChatGPT uses the same URL through its own custom connector flow:

  1. In ChatGPT, open the settings area for connectors (custom connectors / MCP servers).
  2. Add a new connector and paste https://app.sparkletree.io/api/mcp.
  3. Sign in with your SparkleTree account when prompted.

SparkleTree speaks the standard protocol (MCP) that ChatGPT’s connectors use, so this works the same way: paste the URL, sign in, done. Exact menu names may vary as ChatGPT’s interface evolves.

Connect from Cursor, Gemini CLI, Cline, any MCP client

Section titled “Connect from Cursor, Gemini CLI, Cline, any MCP client”

SparkleTree is a standard remote MCP server, so any client that supports remote servers with sign-in connects with the same URL. There is nothing to install; the sign-in happens in your browser the first time a tool is used.

  • Cursor: open Settings → Tools & MCP → Add custom MCP, or add this to ~/.cursor/mcp.json:

    { "mcpServers": { "sparkletree": { "url": "https://app.sparkletree.io/api/mcp" } } }

  • Gemini CLI: run gemini mcp add --transport http sparkletree https://app.sparkletree.io/api/mcp, then /mcp auth sparkletree inside Gemini to sign in.

  • Claude Code: run claude mcp add --transport http sparkletree https://app.sparkletree.io/api/mcp, then /mcp to sign in.

  • Cline: open MCP Servers → Remote Servers, give it a name, paste the URL.

  • Anything else: paste the URL wherever the client accepts a remote or “Streamable HTTP” MCP server. If it offers a transport choice, pick HTTP, not SSE or stdio.

Menu names move around between releases; the URL is the part that matters. Scripts and headless agents that cannot open a browser use a workspace API key instead; see MCP connector setup.

What your AI can do once connected

Section titled “What your AI can do once connected”

Everything happens inside your workspace, limited by your role. Your AI can’t see or touch anyone else’s data. In plain terms, it can:

  • Read your campaigns, products, surfaces and brand book. List them, open one, check what’s live.
  • Create and edit. Draft new campaigns and variants, add products (including importing one straight from a product page URL).
  • Publish and deploy. Put a campaign live on your surfaces.
  • Check analytics. Impressions, reach, clicks, which variant is winning.
  • Render videos. Generate a short on-brand video from a campaign.

So instead of navigating the dashboard, you can say things like “import this product from its URL and draft a campaign for it,” or “summarize last week’s performance and tell me which variant to keep.”

Good to know before you let it loose

Section titled “Good to know before you let it loose”
  • It acts as you. Anything your AI does (publishing, editing, deleting) happens under your account, with your permissions. Treat it like handing your keyboard to a very fast assistant.
  • Reads and writes are separate tools. Reading tools are marked read-only, so most clients run them without asking. Tools that change things or spend credits are not, and a well-behaved client asks first.
  • Generating costs credits. Asking your AI to generate a campaign or render a video uses the same credits as doing it yourself in the dashboard. The prompt below tells assistants to confirm before spending.
  • You can disconnect anytime. Remove the connector in your client’s settings to cut access immediately. If you set up an API key instead, disable it under Settings → API keys in SparkleTree.

Tip: After connecting, give your AI the orientation prompt in Ready prompt for your AI. It teaches the assistant to look before it leaps: read first, confirm before publishing or spending credits.

If something doesn’t work

Section titled “If something doesn’t work”
  • Sign-in loop or “unauthorized”: disconnect the connector and add it again. A fresh sign-in usually fixes it.
  • Your AI says it can’t find SparkleTree tools: start a new conversation after connecting; some clients only pick up new connectors in fresh chats.
  • “Too many requests”: the connection allows 120 requests per minute. Assistants normally stay far below that; wait a minute and try again.
  • No account yet: the connection can’t create one for you. Sign up at app.sparkletree.io first.

Want more?

Section titled “Want more?”