Show HN: Trust infrastructure for AI agents (RFC .well-known autodiscovery)

1 wellknownmcp 0 8/8/2025, 6:06:47 AM wellknownmcp.org ↗
AI agents browse your site but can't verify what they're clicking on. No trust chains, no structured intent, just pixel-guessing.

Built MCP (Model Context Protocol) to fix this:

- RFC 8615 .well-known/mcp.llmfeed.json autodiscovery - Structured JSON feeds declaring site capabilities & intent - Ed25519 cryptographic signatures for trust verification - Cross-LLM compatibility (Claude, GPT, Gemini tested)

Example feed: { "capabilities": [{"intent": "search", "endpoint": "/api/search"}], "trust": {"signed_blocks": ["capabilities"], "algorithm": "Ed25519"} }

Agent workflow: visit yoursite.com → GET /.well-known/mcp.llmfeed.json → discover capabilities → verify signatures → act with context.

30-second test: 1. https://wellknownmcp.org/hn 2. Copy the "Kung Fu" training prompt 3. Feed to any LLM, ask "I know kung fu" 4. Watch them become MCP experts instantly

Real performance vs traditional crawling: 15x faster discovery, 7x cheaper tokens, 95% vs 70% confidence.

Open source, self-hostable, no vendor lock-in. Three-site ecosystem live with independent PKI (LLMCA.org).

Agents need structured intent, not DOM parsing guesswork. It's a draft, a proposal, mixed ideas. Just drop a hot/cold reaction, I need feedback ;)

Comments (0)

No comments yet