Make your domain agent-native.
AI agents don’t browse — they probe. They read llms.txt before your homepage, check .well-known before your docs, and trust what they can verify over what you claim. wellknownmcp is a working collection of doctrine, tools and reference implementations for that reality.
Reference implementation
cortex-gateway
A self-hosted MCP federation gateway — one OAuth 2.1 MCP server in front of N apps, where every agent call carries the real user’s identity. Launched agent-first: llms.txt, an AGENTS.md with probeable claims, RFC 9728 discovery, and a live demo any MCP client can join in ~30 seconds. This is what agent-native shipping looks like.
On this site
- Ecosystem notes — occasional, signed write-ups when something in the agentic-web space actually changes. No content mill.
- Agent-readiness checker — live probes for any domain: MCP handshake, OAuth discovery chain, llms.txt, security.txt, AI-crawler posture. Every result line ships its curl.
- Ecosystem radar — how fast the MCP registry is growing, counted weekly from the official registry API. Machine-generated, no editorial judgment, raw JSON per week.
- Guides — evergreen, practical notes: llms.txt done right, AGENTS.md that’s agent-verifiable, MCP discovery (RFC 9728) in practice. Each ends with a way to verify it yourself.
Research — LLMFeed (2025)
Before MCP won the tool layer and llms.txt won the content layer, we prototyped signed, verifiable agent feeds: LLMFeed. The experiment is archived and documented — the spec and tools remain online as a record, and the questions it asked (trust, signatures, certification) are still open.
Maintained by the wellknownmcp org · everything here is verifiable — probe before you trust.