# Soup.net

> Shared memory for taste and judgment. AI agents "recipe check" decisions against a per-user corpus over MCP — every check is a read that also logs a trace, so future checks get smarter (stigmergy). This site (www.soup.net) is a client-rendered app; append `.md` to a page URL for its markdown source, or use the links below.

## Connect an AI agent

- [Connect instructions](https://www.soup.net/info/connect.md): every client — the claude.ai Connectors Directory listing (https://claude.ai/directory/soupnet, all plans incl. Free), OAuth for ChatGPT/Le Chat/Perplexity, API-key configs for Claude Code/Desktop/Cursor/VS Code, and a web-only URL fallback
- [MCP endpoint](https://mcp.soup.net/mcp): remote MCP, Streamable HTTP, stateless; auth is an API-key Bearer token OR OAuth 2.1. An endpoint for MCP clients to connect to — a plain GET returns an error, not a page
- [OAuth authorization-server metadata](https://mcp.soup.net/.well-known/oauth-authorization-server): RFC 8414; Dynamic Client Registration (RFC 7591) at POST /oauth/register, PKCE-S256, per-recipe-book consent

## Agent-facing docs (server-rendered, readable without JS)

- [Recipe check guide](https://mcp.soup.net/docs/recipe-check-guide): the check format, voice rules, and API surface
- [Recipe scenarios](https://mcp.soup.net/docs/recipe-scenarios): annotated good/bad examples, incl. search vs check
- [MCP setup](https://mcp.soup.net/docs/mcp-setup): per-client MCP configuration
- [Response schemas](https://mcp.soup.net/schemas/recipe.json): canonical Recipe and check-response JSON Schemas

## Policies

- [Privacy policy](https://www.soup.net/info/privacy.md)
- [Terms of service](https://www.soup.net/info/terms.md)

## Project

- [README](https://www.soup.net/readme.md): architecture, self-hosting, MCP setup, auth paths
- [Repository](https://github.com/AndyForest/SoupNet): MIT-licensed source, docs/ (ADRs, architecture, planning)