Setup with an agent
Paste a prompt → the agent does signup, tracking install, Stripe, providers, identify, custom events, first automation.
The prompt
Connect to the FunnelFizz MCP server at https://mcp.funnelfizz.com using
my API key, then read https://docs.funnelfizz.com/skill.md and follow it.
Stop and ask me before anything destructive or anything that needs my credentials.
The agent fetches /skill.md, connects to MCP, and walks the steps. It uses MCP tools for everything that has one (registering tracking sites, validating Stripe keys, polling provider OAuth, verifying email senders by code) and only asks you for things that need a human (signing up, clicking through Stripe's UI, approving code changes).
What the agent does for you
- Fetches
skill.mdand parses it. - Asks the preliminaries: your domain, framework, whether you have Stripe.
- Writes the tracking snippet into your codebase, correctly for your framework.
- Walks you through creating the Stripe restricted key (7 read scopes); validates via MCP.
- Connects providers (X, YouTube, GSC, etc.), kicks off OAuth, polls for completion via MCP.
- Writes
funnelfizz('identify', …)calls into your auth layer. - Instruments custom events at key points.
- Builds a starter trial-welcome automation.
What you still do yourself
| Step | Why |
|---|---|
| Create the FunnelFizz account | Email verification flows are hard for agents |
| Click through Stripe's UI to mint the restricted key | Stripe's dashboard, not ours |
| Sign in to providers during OAuth | The agent kicks it off and waits |
| Approve code changes the agent proposes | Always |
Tips for best results
Give context upfront so the agent doesn't ping-pong with questions:
Next.js 14 App Router with next-auth. Hobby plan. Stripe for billing.
Domain: widgetspro.com, app at app.widgetspro.com.
"Activation" for me means: created first project + invited a teammate.
Ask it to commit along the way so each change is reviewable:
After each major change (tracking install, identify, events), commit
with a clear message. Don't bundle unrelated changes.
Security boundaries
The skill file instructs the agent to never:
- Commit your Stripe key or any FunnelFizz API key to a repo.
- Send tokens / session cookies / keys to third-party services.
- Create Stripe accounts or sub-accounts on your behalf.
If you see your agent about to cross one of these lines, stop it.
If something goes wrong
Tell the agent: "That didn't work, paste the error and walk me through diagnosis. Don't retry blindly." Common failures:
| Symptom | Likely cause |
|---|---|
| Tracking not verifying | Meta tag outside <head> or CSP blocking the script |
| Stripe key rejected | Missing one of the 7 read scopes, error names which |
identify never fires | Script not loaded on authed pages, or auth-state check misses |
| Events missing from splits | Fired before identify, attributed to anonymous visitor |
Writing your own skill file
If you're building agent-friendly product docs, /skill.md is the canonical template. Principles: self-contained (no cross-page fetches), ordered (steps run in sequence), conservative (ask before destructive actions), versioned.