# Agent Friendly Web

> Public website auditor and guided owner workspace created by Gabriel Mucchiut and incubated within Tokenizart.

Agent Friendly Web measures public evidence about how agents can discover, understand, and use a website. Its AF-0 to AF-5 framework is a proprietary methodology published for transparent use; it is not an official certification.

## Primary resources

- [Public auditor](https://agentfriendlyweb.dev/)
- [Methodology](https://agentfriendlyweb.dev/metodologia)
- [Tokenizart case](https://agentfriendlyweb.dev/casos/tokenizart)
- [Agentic maturity demonstrator](https://agentfriendlyweb.dev/evolucion-agentica)
- [Readiness manifest](https://agentfriendlyweb.dev/.well-known/agent-readiness.json)
- [OpenAPI](https://agentfriendlyweb.dev/openapi.json)
- [Public AI catalog](https://agentfriendlyweb.dev/.well-known/ai-catalog.json)
- [Agent Skills index](https://agentfriendlyweb.dev/.well-known/agent-skills/index.json)
- [Security policy](https://agentfriendlyweb.dev/.well-known/security.txt)
- [Public repository](https://github.com/tokenizartinfo-ops/agent-friendly-web)

## Important limits

- Public scans are read-only and do not authenticate to the target website.
- `llms.txt` is a community proposal, not a mandatory crawler standard.
- WebMCP is treated as an experimental W3C Community Group draft.
- The private owner dossier requires sign-in and must not be crawled or inferred.
- MCP, A2A, CLI, WebMCP and x402 are not production capabilities unless the readiness manifest marks them as `deployed` and links a verifiable endpoint.