🧡 Made specifically for Hacker News

Google's AI agents can't verify what they're clicking on.
We built the trust infrastructure they need.

Google Mariner ($249/month), OpenAI Operator, and Anthropic Claude browse autonomously but are trust-blind. Here's the cryptographic verification and structured intent system that makes agents actually useful.

The Problem (30 seconds)

Google Mariner just rolled out to enterprises at $249/month. Early deployments are failing security audits.

Why Visual Agents Fail:

• No verification: Can't tell if site.com is legitimate
• Blind clicking: Guesses button meanings from pixels
• Security gaps: No cryptographic trust chains
• Intent confusion: Misinterprets website purposes

Enterprise Impact:

• Compliance failures: Can't verify data sources
• Fraud vulnerability: No way to detect spoofed sites
• Audit nightmares: Black box decision making
• Liability issues: Who's responsible for agent mistakes?

The Solution: MCP Trust Infrastructure

Make websites declare their intentions with cryptographic proof. Transform agent guessing into agent knowing.

1Agent visits: yoursite.com/.well-known/mcp.llmfeed.json

2Discovers: Structured capabilities, auth methods, business intent

3Verifies: Cryptographic signatures (if present)

4Acts: With full context and verified trust

Try This Right Now (2 minutes)

Pro tip: Train your LLM first with our universal prompt for instant expertise

The "Kung Fu" Test:

🌐 Quick Demo Script🔍 Live

Ready to test

# Step 1: Train your LLM
Go to https://wellknownmcp.org/train and copy the universal training prompt

# Step 2: Feed the training to any LLM (ChatGPT, Claude, Gemini)
[Paste the training prompt]

# Step 3: Test comprehension 
Ask your LLM: "I know kung fu"

# Result: Your LLM becomes an instant MCP expert

Basic MCP Feed:

🌐 Sample MCP Implementation🔍 Live

Production-ready example

{
  "feed_type": "mcp",
  "metadata": {
    "title": "My Site Agent Interface",
    "description": "What agents need to know about this site"
  },
  "capabilities": [
    {
      "intent": "search_products",
      "method": "GET",
      "endpoint": "/api/search",
      "trust_level": "public"
    }
  ],
  "trust": {
    "signed_blocks": ["capabilities"],
    "algorithm": "Ed25519",
    "scope": "partial"
  }
}

Live Implementation

See this in action: wellknownmcp.org/.well-known/ (RFC 8615 compliant)

Core Feed Navigation Capabilities

Built on Real Standards

Cryptographic Trust

Ed25519 signatures, PKI infrastructure, certificate chains

RFC Standards

RFC 8615 .well-known/, OAuth 2.0, JSON Schema validation

Enterprise Ready

Audit trails, compliance logs, governance controls

Working Ecosystem (3 Live Sites)

WellKnownMCP.org

Specification & Tools

LLMCA.org

Certification Authority

LLMFeedForge.org

Visual Builder (in dev)

Real Performance Comparison

We tested Claude analyzing OpenAI.com vs WellKnownMCP.org. Results speak for themselves:

Metric	OpenAI.com (Traditional)	WellKnownMCP.org (MCP)	Advantage
Discovery Time	30+ seconds	2 seconds	15x faster
Token Cost	~15,000 tokens	~2,000 tokens	7.5x cheaper
Confidence Level	70%	95%	25% more reliable
Trust Verification	None	Cryptographic	Security upgrade