# cuppy.io

> Documentation hubs, governed MCP services and downloadable skill packages that let an AI agent work
> with a software product safely. Every public page and every public document is open to anonymous
> crawlers and anonymous agents; no credential is needed to read.

This file is the SITE directory, in the llmstxt.org form: it names the pages and the entry points of
cuppy.io itself. It does not list documents. Each hub publishes its OWN catalog at
`https://cuppy.io/{hub}/llms.txt`, which lists every public document in that hub with its folder, role,
shape, short id and size — that is the file to read when you want content rather than navigation.

## Agent entry points

- [MCP endpoint](https://cuppy.io/mcp): the Model Context Protocol endpoint. Call `initialize`, then `tools/list`, then `_hello` for a boot screen that names what to read first. Anonymous reads are served; a write needs a credential.
- [Hub directory, machine-readable](https://cuppy.io/.well-known/agents/providers.json): one entry per hub, with its host and its `/mcp` URL.
- [Signed agent manifest](https://cuppy.io/.well-known/agents/manifest.jws): the provider's manifest, with its verify key at `/.well-known/agents/jwks.json`.

## Document catalogs, per hub

- [cuppy hub catalog](https://cuppy.io/cuppy/llms.txt): the platform's own documentation — the hub standard, the served surface, the agent-ergonomics guide and the boot file.
- [Hub directory](https://cuppy.io/hubs): every hub this platform serves. A hub's catalog is always `/{hub}/llms.txt` and its documents are always `/{hub}/docs/{path}.md` as raw Markdown.

## Pages

- [Home](https://cuppy.io/): what the platform is and who it is for.
- [Services](https://cuppy.io/services): the directory of hosted agent services, by industry.
- [Skills](https://cuppy.io/skills): public skill packages — search and download versioned expertise bundles.
- [Pricing](https://cuppy.io/pricing): pricing for producers and for developers.
- [Architecture](https://cuppy.io/architecture): how agent enablement works here.

## Report a problem

- [MCP endpoint](https://cuppy.io/mcp): something wrong in a document? Call the `session feedback` tool — `kind` is one of `bug`, `stale_doc`, `standard_gap` or `request`, with a `title`, a `body` and the document's `path`. No credential is needed, and the hub's maintainer reads it.

## Machine-readable

- [Sitemap](https://cuppy.io/sitemap.xml): the site's pages. A hub's documents are in its own sitemap at `/{hub}/sitemap.xml`.
- [robots.txt](https://cuppy.io/robots.txt): the crawl rules, and the sitemap pointer.