Docs for AI

If you're asking an AI assistant to write code against Kustomer, give it these docs rather than letting it rely on what it remembers. Everything here is available in plain formats that fit in a context window and stay current with the site.

llms.txt

/llms.txt is a Markdown index: a summary of the platform, then a link and description for every docs page grouped by section, then the API references. It's small enough to paste into any assistant, which can then fetch only the pages it needs.

/llms-full.txt concatenates every docs page. Use it when your tool can't fetch URLs, or to load the docs into a project knowledge base. It doesn't include the per-operation API reference; for that, use the API Markdown.

Markdown versions

Add .md to any docs page URL to get its Markdown source, rendered for reading by a model:

curl https://developer.kustomer.com/docs/apps/app-definition.md
curl https://developer.kustomer.com/api/core-resources/get-customer-by-email.md

• Components become standard Markdown: callouts become blockquotes, field lists become bulleted lists, tabs become labeled sections, and endpoint badges become the method and path with a link to the reference.
• Every interactive diagram on the site has a written description, and the Markdown version includes it in place of the animation. Mermaid diagrams stay as Mermaid source.
• Code samples are kept with their language tags.

Each HTML page links its Markdown version in its structured data (below), so crawlers can find it.

Page actions

Every docs page has a Copy page button that copies its Markdown, and a menu to:

• View as Markdown: opens the .md version.
• Open in ChatGPT, Open in Claude, or Open in Cursor: starts a conversation with a prompt that points the assistant at the page, so you can ask questions or build from it.

Operation index

To give an agent the API as tools, use the kusty MCP server.

/api/operations.json lists every operation with its operationId, method, path, tag and reference URL, for tools that need to map an operationId to its documentation:

{ "spec": "core-resources", "slug": "get-customer-by-email", "operationId": "GetCustomerbyEmail", "method": "GET", "path": "/v1/customers/email={email}", "summary": "Get customer by email", "tag": "Customers", "deprecated": false, "url": "https://developer.kustomer.com/api/core-resources/get-customer-by-email" }

Docs MCP server

The docs MCP server lets an assistant search and read these docs as tools, instead of fetching pages itself. It's read-only and needs no Kustomer credentials. The kusty MCP server is the one that calls the API.

It runs over stdio with Bun from a checkout of the docs source, and reads from developer.kustomer.com, so there's nothing to build:

{
  "mcpServers": {
    "kustomer-docs": {
      "command": "bun",
      "args": ["/path/to/kustomer-developer-docs/mcp/server.ts"]
    }
  }
}

Set KUSTOMER_DOCS_URL to read from another deployment. In a checkout that has run bun run prebuild, it reads the local build instead.

Structured data

Every page includes TechArticle (docs) or APIReference (operations) JSON-LD with its title, description, section and canonical URL, and an encoding entry pointing at its Markdown version:

{
  "@context": "https://schema.org",
  "@type": "TechArticle",
  "headline": "App definition",
  "url": "https://developer.kustomer.com/docs/apps/app-definition",
  "encoding": { "@type": "MediaObject", "encodingFormat": "text/markdown", "contentUrl": "https://developer.kustomer.com/docs/apps/app-definition.md" }
}

Prompting tips

1. Pin the docs, not the model's memory

   Older Kustomer examples on the web use retired endpoints, deprecated SDK calls and outdated versions. Tell the assistant to use these docs: "Use https://developer.kustomer.com/llms.txt as the source of truth for the Kustomer API."

2. Give it the exact page

   For a focused task, paste the URL of the page plus .md, such as the operation it should call. A single page is more reliable than the whole corpus.

3. Let it check the request

   Ask the assistant to validate request bodies against the operation's schema in the API reference, or use kusty's --dry-run and schema validation to check its requests before they run.