VipterHelp Center

Integrating with AI agents

What Vipter publishes so that an AI agent (Claude Code, Cursor, Codex, ChatGPT and the like) can integrate the API without human help: the vipter-api skill in the Agent Skills format, llms.txt with instructions, every documentation page in Markdown, the OpenAPI 3.1 document and the Postman collection; how to point the agent at each one, what to ask it, and what never to hand it.

Admin or OwnerAll plans

A large share of integrations today is written by an AI agent from a request in plain language. Vipter publishes what that agent needs to read to get it right the first time: the API rules in a skill file, an index of pages in llms.txt, every page in Markdown, and the API description as OpenAPI and as a Postman collection. This page says where each piece is and how to use it.

What is published

PieceAddressWhat it is for
vipter-api skillhttps://docs.vipter.com/skills/vipter-api/SKILL.mdInstructions in the Agent Skills format (frontmatter name and description, body in Markdown): authentication, conventions, objects and their Stripe counterparts, step-by-step workflows, pitfalls. It is the first file the agent should read.
Skills indexhttps://docs.vipter.com/skills/index.jsonLists the published skills, with version and links.
llms.txthttps://docs.vipter.com/llms.txtA summary for agents at the top and the list of every page, in three languages, with the link to each one's Markdown.
Pages in Markdownany page of this help center with .md at the endExample: https://docs.vipter.com/en/developers/api-reference.md. No menu, no HTML, just the content.
OpenAPI 3.1https://api.vipter.com/v1/openapi.jsonEvery endpoint, parameter and object, with schemas. For generating clients and for the agent to check fields.
Postman collectionhttps://api.vipter.com/v1/postman.jsonGenerated from the OpenAPI document on every call: one folder per resource, authentication through the apiKey variable, example bodies. Imports into Postman, Bruno and Insomnia.

None of this needs a key. The key only comes in when the agent runs the code it wrote.

How to point the agent

Install the skill in the project, so it is loaded whenever the task mentions Vipter:

mkdir -p .claude/skills/vipter-api
curl -sL https://docs.vipter.com/skills/vipter-api/SKILL.md -o .claude/skills/vipter-api/SKILL.md

Then ask for the integration in plain language:

Integrate our app's checkout with Vipter: when the user clicks "subscribe", create a
checkout session with the user's ID, redirect, and activate the plan on the
checkout.session.completed webhook. The key is in VIPTER_API_KEY. Use the vipter-api skill.

What to ask the agent

Requests the skill covers well, by frequency:

  1. Checkout with the user identified: checkout session with client_reference_id, redirect, confirmation by GET and by webhook. The human guide is SaaS: from sign-up to dashboard.
  2. Off-cycle charge on the subscription: POST /v1/subscriptions/{id}/charges with Idempotency-Key, handling of the 402.
  3. Metered usage: create a meter, report events, price it on the offer or on the subscription.
  4. Webhooks: endpoint, signature verification, deduplication, switch by event type.
  5. Migration from Stripe: swap the code fragments listed in Migrating from Stripe to Vipter while keeping both providers during the cut-over.

Always ask that the code read the key from an environment variable and confirm the key with GET /v1/account before any other call. The skill already says so; saying it again costs nothing.

What never to hand the agent

  • The vk_live_… key in the prompt. The agent does not need it to write the code; it needs it to run it. Put the key in the project's .env and let the agent read the variable's name, not its value. If a key leaked into a chat, revoke it and create another.
  • The endpoint's whsec_… secret, for the same reason.
  • Permission to run charges in production without a test connection. Vipter has no test mode; the provider's test connection is what separates a test from a real charge. See Testing without a test mode.

Checking what the agent did

  • The request logs in the Developers tab show every call the key made, with status and duration.
  • The deliveries of each endpoint show whether the agent's server answered 2xx.
  • GET /v1/events lists what happened in the store, in the endpoint's catalog.

What comes next

A remote MCP server, for the agent to query and act on the store without writing code, and a thin Node.js SDK are planned, with no date. The skill, the OpenAPI document and the Postman collection already cover integration by code.

What to do next

Was this page helpful?

On this page

Language