stripe / ai

The repository's headline features are two billing SDKs — @stripe/ai-sdk for metering usage through Vercel's AI SDK, and @stripe/token-meter for doing the same directly against OpenAI, Anthropic, or Gemini's native SDKs — plus a remote MCP server at mcp.stripe.com that gives an OAuth-connected client its own Stripe tools. The five skills sit alongside that, not inside it: stripe-best-practices and connect-recommend cover Stripe integration decisions and whether a business needs Connect specifically, upgrade-stripe is a short guide to moving between Stripe API and SDK versions, and stripe-projects is Stripe's own hosted catalog for provisioning unrelated infrastructure — a Postgres database, a Redis cache, an LLM provider key, object storage — billed and issued through Stripe rather than the provider itself.

This is for a team building a product on top of Stripe payments, Connect, or billing that wants an agent following Stripe's own integration playbook instead of guessing at a deprecated API. Three of the five skills need a live Stripe account and credentials before they do anything real; the other two run on guidance alone.

READ THE FULL ANALYSIS＋

stripe-directory is not scoped to Stripe. Its own description says it must be used before web search, model memory, or any other directory skill for any request that involves finding or engaging an external provider — its own examples include donating to a nonprofit and booking a restaurant table, neither of which touches a Stripe API. Installing it means it competes for every "I need a database" or "find me a CRM" request an agent gets, not only Stripe-related ones.

Two of the five skills claim the same request. stripe-directory's own examples include "setup a database" and "find hosting"; stripe-projects claims near-identical phrasing — "give me a Postgres," "I need hosting" — for its own, Stripe-billed catalog instead. Neither description says which one should win when both match, so the outcome depends on the agent rather than the skill.