Originally published at devtoolpicks.com
Your AI coding agent can now create a Cloudflare account, register a domain, start a paid subscription, and deploy a live application. All from one CLI session. No dashboard. No copy-pasting API keys. No manually entering a credit card.
This is not a demo. Stripe Projects shipped this in open beta at Stripe Sessions 2026, and thousands of developers joined the waitlist in the first two days. The system works through a protocol that Cloudflare and Stripe co-designed, and it is already integrated with providers like Supabase, Vercel, PostHog, Clerk, Neon, PlanetScale, Twilio, and Hugging Face.
If you build with AI coding agents, this changes how deployment works. Here is what is actually happening, how the protocol works, and what guardrails you need before letting an agent spend money on your behalf.
What Stripe Projects actually does
Stripe Projects is a CLI tool that lets you (or your agent) provision your entire dev stack from the terminal. Hosting, databases, authentication, analytics, AI services, observability. One command per service, credentials synced to your local environment, billing handled through Stripe.
Before this, deploying a side project meant signing up for 4-5 services separately, managing API keys across dashboards, entering payment details on each one, and hoping you remembered to set up billing alerts. An AI agent could write the code but it hit a wall the moment it needed to create an account or pay for something.
Stripe Projects removes that wall. The agent calls stripe projects add cloudflare/registrar:domain and the system handles discovery, account creation, payment, and credential storage in one shot.
How the protocol works
The Cloudflare-Stripe protocol has three components that run in sequence.
Discovery. The agent calls stripe projects catalog and gets back a JSON list of every available service. Hosting, databases, auth providers, analytics tools, AI services. The agent picks what it needs based on your prompt. You do not need to know which providers exist or what they offer. The catalog is the context.
Authorization. When the agent provisions a service, Stripe acts as the identity provider. If you already have an account with the provider (say, Cloudflare), you go through a standard OAuth flow to grant access. If you do not have an account, the provider creates one automatically using your Stripe identity. No signup forms. No email verification loops.
Payment. Stripe provides a tokenized payment method to the provider. Your actual credit card number never reaches the agent or the provider directly. Stripe sets a default spending cap of $100 per month per provider. The agent cannot exceed that limit without you explicitly raising it.
That last point matters. If you read about AI agents burning through bills overnight, the $100 cap is Stripe's direct answer to that problem. It is enforced at the API level, not as a suggestion.
The provider ecosystem
Stripe Projects launched with a catalog that covers most of what an indie hacker needs to ship a product.
Hosting: Cloudflare Workers, Vercel, Railway
Databases: Supabase, Neon, PlanetScale, Turso
Auth: Clerk
Analytics and observability: PostHog
AI: Hugging Face, Chroma
Communications: Twilio
Other: AgentMail, RunloopAI
The list is growing. Stripe is actively onboarding providers through a partner intake process. Any SaaS or infrastructure company that bills through Stripe can potentially integrate.
For indie hackers, the most useful combination is probably Cloudflare (hosting + domain) plus Supabase (database + auth) plus PostHog (analytics). That gives you a full production stack, provisioned from one terminal session, with billing consolidated in your existing Stripe account.
If you already use Vercel or Hetzner for hosting, Vercel is in the catalog. Hetzner is not (yet). Railway is also available as an option.
What this looks like in practice
Here is the actual workflow. Install the Stripe CLI with the Projects plugin. Log in to Stripe. Start a new project.
stripe projects init my-saas-app
stripe projects add cloudflare/workers:site
stripe projects add cloudflare/registrar:domain
stripe projects add supabase/database:postgres
stripe projects add clerk/auth:user-management
stripe projects add posthog/analytics:events
Each command provisions a real resource in your own account with that provider. Credentials are stored in an encrypted vault and synced to your local .env file. The agent can then use those credentials to deploy code, run migrations, and configure services.
The entire flow from zero (no accounts, no domains, no database) to a live production app can happen in a single agent session. Stripe calls it going from "literal zero" to deployment.
The guardrails you need
The $100/month default cap is a solid starting point, but it is not enough on its own. Here are the failure modes to watch for.
Wrong domain purchase. Your agent reads a vague spec and buys myapp-staging.io when you meant myapp.io. Domain purchases are non-refundable. Be specific in your prompts about exact domain names.
Deploying to the wrong account. If the agent has credentials for multiple Cloudflare or Vercel accounts (personal and work, for example), it might pick the wrong one. Review your stripe projects status output before the agent starts deploying.
Retry loop on paid services. An agent hits a flaky API, retries the call, and each retry triggers a metered charge. At machine speed, this can burn through the $100 cap in minutes. Set up budget alerts on every provider, not just the Stripe-level cap.
The $100 cap is per provider, not total. If you have 5 providers connected, an agent could theoretically spend $500/month before any cap kicks in. Know your total exposure.
The practical checklist before giving your agent access to Stripe Projects: set the spending cap to something you are comfortable losing, enable budget alerts on every provider, never let the agent run unattended on the first session, and check stripe projects status after every deployment.
Why this matters for indie hackers
If you ship solo, you are probably running 5-10 SaaS subscriptions with separate logins, separate billing, and separate dashboards. Stripe Projects consolidates that into one CLI with one billing relationship.
The bigger shift is what happens when agents can handle the entire deployment lifecycle. Right now, most developers use Claude Code or Cursor to write the code and then switch to manual mode for deployment, DNS, and billing. That manual step is where projects stall. You write the code in 20 minutes and then spend an hour configuring the infrastructure.
Stripe Projects compresses that hour into a few CLI commands that your agent can run. For indie hackers building side projects, that is the difference between shipping on a Saturday afternoon and shipping next month.
The protocol is also open by design. Any platform with signed-in users can act as the orchestrator in the same way Stripe does. Cloudflare has already said they want more platforms to integrate. If this takes off, the catalog of services available to agents will grow fast.
The catch
Stripe Projects is in open beta. That means the provider list is still limited, some integrations are rough around the edges, and the protocol specification is not yet finalized.
The payment flow (Shared Payment Tokens) is in developer preview only in the US, Europe, UK, and Canada. If you are outside those regions, you can still use the provisioning and credential management features, but upgrading to paid tiers through the CLI is not available yet.
And the security model, while thoughtful, is new. The $100 cap and payment tokenization are good defaults. But this is the first major system that gives AI agents real purchasing authority at scale. The attack surface for things like token fraud and agent impersonation is still being mapped out. Stripe expanded Radar to address some of these risks, but it is early days.
FAQ
Is Stripe Projects free?
The CLI tool itself is free. You pay for the services you provision through it (Cloudflare, Supabase, PostHog, etc.) at their normal rates. Stripe does not add a surcharge. Most providers have free tiers, so you can set up a full stack at $0 if you stay within those limits.
Do I need a Stripe account to use it?
Yes. You need to be signed in to Stripe via the CLI. Stripe acts as the identity provider and payment handler for the entire system. If you already accept payments through Stripe for your SaaS, you can use the same account.
Can I use this without an AI agent?
Yes. Stripe Projects works as a regular CLI tool. You can run all the commands manually. The agent integration is an extension, not a requirement.
What happens if I exceed the $100 spending cap?
The cap is enforced at the API level. The agent's request is rejected. It does not get a warning and proceed anyway. You can raise the limit in your Stripe dashboard if you need higher spending authority for a specific provider.
Which coding agents work with this?
Any agent that can run CLI commands. The Stripe docs specifically mention Claude Code and similar tools. When you initialize a project, Stripe writes agent skills into your local directory so the agent has context about what commands are available.
Bottom line
Stripe Projects is the most significant infrastructure announcement for indie hackers in 2026 so far. Not because any individual feature is new, but because it solves the boring problem that actually blocks shipping: the gap between "the code is done" and "the app is live."
AI agents writing code is already normal. AI agents deploying code, buying domains, and paying for infrastructure is new. Stripe and Cloudflare just made it real.
The $100/month default cap, the payment tokenization, and the provider catalog are well-designed. But set up your guardrails before you let an agent loose. Check your spending limits. Enable alerts. And run the first session manually before going fully autonomous.
Stripe Projects is at projects.dev. Install the CLI, connect a provider, and try it.
Top comments (0)