How it works

Three steps from zero to fully stacked.

Claude audits your environment, AI agents provision every account, humans handle the blockers, and credentials land in your encrypted vault — all within 48 hours.

The audit

Claude scans your project locally, detects what accounts you already have, and sends the results to Stacked to see exactly what's missing and what it costs — before anything is created.

1. Install the Stacked MCP server

Add the Stacked MCP server to your Claude Code config. This gives Claude access to tools for auditing your stack, placing orders, and retrieving credentials. You only need to do this once.

Get your API key from your account dashboard, then add the server config shown on the right.

2. Open your project and ask Claude

Navigate to the project you want to audit, then tell Claude to run the audit. Claude scans files locally on your machine — nothing is uploaded. It reads your .env files, package.json, config files, and source imports to detect which services you already use.

If you have multiple projects, just cd into the one you want audited before asking. Claude always audits the project in your current working directory.

What to say to Claude
$ cd ~/projects/my-saas-app
$ claude
"Audit my stack and tell me what accounts I need to launch."

3. Claude scans, Stacked prices

Claude reads your project files to build a list of services you already have (Supabase, Vercel, GitHub, etc.), then calls the Stacked audit_stack tool with that list. Stacked compares it against the full SaaS launch catalog and returns exactly what's missing, with pricing per account.

  • Active accounts detected via .env files, config references, and SDK imports
  • Missing accounts required by your dependencies (e.g., Stripe SDK installed but no API key)
  • Deployment target gaps (Vercel, Netlify, Cloudflare)
  • Social and distribution accounts your product type typically needs
  • Per-account pricing with standard and same-day SLA options

You review the results and approve before anything is ordered. Claude will ask you to confirm and handle payment before any accounts are created.

~/.claude/claude_desktop_config.json
{
  "mcpServers": {
    "stacked": {
      "command": "node",
      "args": ["/path/to/stacked-help/mcp/server.js"],
      "env": {
        "STACKED_API_KEY": "sk_stacked_your_key"
      }
    }
  }
}
What Claude does under the hood
// 1. Claude scans your project locally
//    Reads .env, package.json, vercel.json, etc.
//    Detects: supabase, vercel (already configured)

// 2. Claude calls Stacked's audit_stack tool
audit_stack({
  "existing": ["supabase", "vercel"],
  "tier": "auto"
})

// 3. Stacked returns the gap analysis
{
  "existing": [
    { "name": "Supabase" },
    { "name": "Vercel" }
  ],
  "missing": [
    { "name": "Stripe", "price": "$6" },
    { "name": "Resend", "price": "$4" },
    { "name": "Sentry", "price": "$5" }
  ],
  "total": "$15"
}
Multiple projects?
Claude audits whichever directory you're in when you start it. Switch projects with cd ~/projects/other-app and run the audit again.

The provisioning

AI agents attempt every signup automatically. When they get blocked, a human agent takes over instantly with full context of where the AI stopped.

AI agents go first

The moment you approve your order, AI browser agents begin working through every account signup in parallel. Each agent navigates the real signup flow — filling forms, verifying emails, accepting terms, and capturing credentials.

For straightforward services like Resend, Supabase, or Vercel, the AI completes the entire process end-to-end in under five minutes. No human involvement needed.

Humans handle the blockers

Some services throw barriers that AI cannot bypass: CAPTCHAs, phone verification, identity checks, or unusual UI patterns. When the AI agent hits one of these, the job is immediately escalated to a human agent.

The human agent receives full context of the attempt:

  • A screenshot of exactly where the AI got stuck
  • The last completed step in the signup flow
  • Any partial data already entered (email, username selected)
  • The specific blocker type (CAPTCHA, SMS, identity, other)

There is no re-work. The human picks up precisely where the AI left off. Most escalated accounts are completed within 15 minutes of handoff.

Speed and accountability

Every provisioning session is recorded. You can request the session log for any account. Most full stacks (5–10 accounts) are completed in under an hour. Complex stacks with App Store or identity-verified accounts are done within 24 hours.

AI agent: automated signup
Browser automation navigates real signup flows. Form fills, email verification, terms acceptance — all handled programmatically. Avg completion: 3–5 minutes per account.
Blocker detected
CAPTCHA, phone verification, identity check, or unusual UI. AI pauses immediately and captures current state: screenshot, last step, partial data.
Human agent takes over
Verified human agent receives full context. Picks up at the exact step the AI stopped. No re-work. Avg resolution: under 15 minutes.
Credential captured
Login URL, username, password, API keys, and any tokens — all captured and encrypted. Account marked complete. Bounty released after 48-hour quality hold.

The vault

Every credential is delivered to your encrypted vault, pre-formatted for 1Password CLI. Pull any secret into Claude Code with a single op:// reference.

Encrypted vault delivery

Once every account in your order is marked complete, your credentials are packaged into a single encrypted vault payload. The vault is accessible from your Stacked dashboard and can be imported directly into 1Password.

Pre-formatted for 1Password CLI

Every credential is structured as a 1Password item with the correct field types: login URLs, usernames, passwords, API keys, and secret tokens. No manual entry. One import command and your entire stack is in 1Password.

op:// references in Claude Code

With 1Password CLI configured, Claude Code can pull any credential using an op:// reference. Instead of hardcoding API keys in your .env file, reference them securely:

# .env (secure references, no plaintext)
STRIPE_SECRET_KEY=op://stacked-vault/stripe/secret-key
RESEND_API_KEY=op://stacked-vault/resend/api-key
SUPABASE_URL=op://stacked-vault/supabase/url
ANTHROPIC_API_KEY=op://stacked-vault/anthropic/api-key

Access revoked after delivery

After your vault is delivered, all Stacked access to your accounts is revoked. Credentials are purged from our systems. You are the sole owner from the moment of handoff. Every provisioning session is recorded and available on request for your audit trail.

  • Accounts created in your name, with your email, on your billing
  • No credential retention on Stacked servers post-delivery
  • Session recordings available for compliance review
  • You can rotate every credential after import to verify full ownership
Your vault
Encrypted
Stripe op://stacked-vault/stripe
Resend op://stacked-vault/resend
Anthropic op://stacked-vault/anthropic
GitHub op://stacked-vault/github
Vercel op://stacked-vault/vercel
5 accounts · op:// ready 1Password CLI
Import to 1Password
# Import your full vault in one command
$ op item create --template stacked-vault.json

# Verify import
$ op item list --vault stacked-vault
ID          TITLE       CATEGORY
stripe      Stripe      Login
resend      Resend      API Credential
anthropic   Anthropic   API Credential
github      GitHub      Login
vercel      Vercel      Login

The full provisioning flow

From your first audit_stack call to credentials in your vault — every step, every handoff, every outcome.

Start
Customer
Connects MCP
Audit
MCP audit
audit_stack()
Order
Order created
create_order()
Execute
AI agent
Browser automation
Decision
Blocker?
CAPTCHA / SMS / ID
Stored
Credential captured
Encrypted
Deliver
Vault delivered
get_vault()
Done
Customer
op:// ready
No blocker
AI completes signup.
Credential captured
automatically.
Blocker hit
Human agent notified.
Picks up at exact step.
Resolves in ~15 min.

Ready to audit your stack?

Create an account, generate an API key, install the MCP server, then open your project and ask Claude to audit your stack.

Create account → Install MCP server →