Stripe Projects Goes Live: Cloudflare Becomes First Cloud AI Agents Can Buy Without a Human
Stripe and Cloudflare have published an open protocol that lets AI agents create cloud accounts, register domains, start paid subscriptions, and deploy applications to production without a human ever opening a dashboard or pasting an API token. The companion product, Stripe Projects, entered open beta this week with a default cap of $100 per month per provider and $100,000 in credits available to Stripe Atlas startups. The launch turns Stripe into the de facto identity and billing layer for autonomous developer agents, with Cloudflare as the first integration partner.
The protocol is built on three components: a discovery endpoint that returns a JSON service catalog, OAuth-based authorization through Stripe as the identity provider, and tokenized payment that never exposes raw card numbers to the agent. Cloudflare engineers Sid Chatterjee and Brendan Irvine-Broque framed the problem bluntly in their announcement.
“Coding agents are great at building software. But to deploy to production they need three things from the cloud they want to host their app: an account, a way to pay, and an API token.”
— Sid Chatterjee and Brendan Irvine-Broque, Cloudflare
How Stripe Projects Works
A developer installs the Stripe CLI with the Projects plugin, signs in once, and runs stripe projects init. From that point the agent can request a Cloudflare account, provision DNS, mint SSL certificates, and ship code without further prompts. Stripe automatically creates the destination account if one does not exist, so the user never visits Cloudflare’s signup form.
Four human approval gates remain mandatory: initial Stripe authentication, terms-of-service acceptance, billing setup, and the merge decision before code reaches production. Everything between those gates is automated. Cloudflare confirmed integrations with AgentMail, Supabase, Hugging Face, Twilio, and more than 70 other providers at launch.
Machine Payments Protocol Under the Hood
The Stripe Projects authorization flow rides on the Machine Payments Protocol (MPP), co-authored by Stripe and Tempo Labs and first published on March 18. MPP itself sits on top of x402, the HTTP 402 Payment Required status code revived as a money-movement primitive. The protocol is rail-agnostic, supporting cards, stablecoins, and bank transfers, and was designed to be extensible by any platform with signed-in users. Visa endorsed MPP earlier in the spring through its Acceptance Platform, releasing a card-based specification and an SDK that integrates Visa Intelligent Commerce and Trusted Agent Protocol capabilities.
For the broader payments market, the launch puts Stripe in a position no rival currently matches. PayPal and Adyen have both announced agent-payment roadmaps for 2026, but neither has shipped a working OAuth-plus-tokenization handshake with a major cloud provider. Square has stayed out of the autonomous-agent conversation entirely, focusing on point-of-sale extensions. Stripe processed $1.9 trillion in total payment volume during 2025, up 34 percent year over year, and now powers 5 million businesses including 90 percent of the Dow Jones Industrial Average.
Why the $100 Cap Matters
The default $100-per-month spending cap is the most consequential design decision in the entire protocol. It forces agents to operate within a budget that a human can rationalize without weekly oversight, while still being generous enough to cover a typical Cloudflare Workers deployment, a domain registration, and a small Twilio or Supabase tier. Developers can raise the limit per provider, but the default keeps a runaway loop from draining a credit card before anyone notices.
This pattern mirrors how marketplaces handle agentic storefronts in dedicated admin sections: explicit channel boundaries plus measurable spend caps. The difference is that Stripe Projects pushes the same model down into infrastructure, where the failure mode is not a bad ad placement but an over-provisioned server farm.
Stripe’s Identity-Provider Play
The strategic move worth watching is Stripe’s expansion from payment processor to identity provider. By acting as the OAuth authority for Cloudflare and the 70-plus partner platforms, Stripe is positioning itself as the login layer of choice for agent-driven workflows. Every developer who uses Stripe Projects is implicitly using Stripe SSO for their entire toolchain. This is the same playbook Plaid used to colonize bank-account verification, and Stripe is executing it three years earlier than most analysts expected.
Stripe’s subscription business already handles the billing data for thousands of SaaS companies, but Projects pulls the company upstream of the purchase decision. When an agent picks Cloudflare over AWS because the integration is friction-free, the choice is functionally a Stripe-default rather than a vendor evaluation. AWS, Azure, and Google Cloud currently have no comparable agent-provisioning protocol on the market.
What Marketers Should Watch
Agent-driven infrastructure spending will show up in two analytics signals this year. First, organic traffic from referrer strings tied to coding agents (Claude Code, Cursor, GitHub Copilot Workspace) will start landing on signup pages of any vendor that integrates with Stripe Projects. Second, attribution models will need to handle a new actor: the agent itself, which sits between the human decision and the recorded conversion. Marketing teams that still attribute the entire funnel to the last human touchpoint will misread the data.
Stripe has not published a public timeline for general availability, but the open-beta status suggests the company wants real production traffic before locking the spec. Cloudflare confirmed the protocol is open and that any platform with signed-in users can integrate the same way — a deliberate signal that Stripe Projects is intended as a standard, not a walled garden.
