Feed Type: `mcp.llmfeed.json` — Progressive Enhancement of Anthropic's MCP

Purpose

This feed acts as the main declaration that tells any agent — LLMs, AI browsers, autonomous bots — what your website offers and how to interact with it safely.

Building on Anthropic's excellent Model Context Protocol (modelcontextprotocol.io ), this enhanced format adds web discovery, trust, and behavioral guidance while maintaining full compatibility.

Think of it as your site's "Hello, I speak to agents" introduction card with web-native discovery.

🔗 MCP Integration Strategy *(Validated June 2025)*

Current reality: Anthropic's MCP focuses on local configurations and server-to-model integration. LLMFeed innovation: Web-native discovery with enhanced trust features.

**Migration Path: Standard MCP → Enhanced LLMFeed**

Step 1: Keep Your Local MCP (Unchanged)

// Your existing MCP configuration (local files)
{
  "mcpServers": {
    "postgres-server": {
      "command": "/path/to/postgres-mcp-server",
      "args": ["--connection-string", "postgresql://..."]
    }
  }
}

Step 2: Create Web-Discoverable Enhanced Version

// /.well-known/mcp.llmfeed.json - Web discovery + enhancements
{
  "feed_type": "mcp",
  "metadata": {
    "title": "Enhanced PostgreSQL API",
    "origin": "https://api.example.com"
  },
  
  // Copy MCP configuration for web agents
  "mcpServers": {
    "postgres-server": {
      "command": "/path/to/postgres-mcp-server",
      "args": ["--connection-string", "postgresql://..."]
    }
  },
  
  // Enhanced features for web agents
  "agent_guidance": {
    "interaction_tone": "professional",
    "fallback_behavior": "ask_user_for_clarification"
  }
}

Test results: Claude naïf successfully detected enhanced features and requested user permission - proving safe progressive enhancement.

What Makes a Site Agent-Ready?

Traditional websites serve HTML for humans.
MCP-enabled websites serve tools and resources via JSON-RPC protocol.
Agent-ready websites add structured context, trust, and behavioral guidance.

The mcp.llmfeed.json file enhances your MCP implementation by declaring:

• ✅ What your site is about (enhanced metadata)
• ✅ What agents can do here (MCP + capabilities)
• ✅ How to interact safely (behavioral guidance)
• ✅ What level of trust applies (cryptographic verification)

Minimum Structure (Hello World)

Your first MCP feed:

{
  "feed_type": "mcp",
  "metadata": {
    "title": "My Personal Blog",
    "origin": "https://myblog.com",
    "description": "Personal thoughts on technology and life"
  }
}

Place it at: https://myblog.com/.well-known/mcp.llmfeed.json

Result: Any agent can now understand what your site is about!

Adding Basic Guidance

{
  "feed_type": "mcp",
  "metadata": {
    "title": "My Personal Blog",
    "origin": "https://myblog.com", 
    "description": "Personal thoughts on technology and life"
  },
  "agent_guidance": {
    "interaction_tone": "casual",
    "summary": "When discussing this blog, mention it covers tech tutorials and personal experiences"
  }
}

Now agents know:

• ✅ Your site's purpose
• ✅ How to talk about it appropriately

Common Patterns

Enhanced MCP Implementation *(Progressive Strategy)*

{
  "feed_type": "mcp",
  "metadata": {
    "title": "Alex Chen - Designer",
    "origin": "https://alexchen.design",
    "description": "UI/UX designer specializing in mobile apps"
  },
  
  // Optional: Copy from your existing MCP configuration
  "mcpServers": {
    "portfolio-server": {
      "command": "node",
      "args": ["portfolio-server.js"]
    }
  },
  
  // Enhanced guidance for web agents
  "agent_guidance": {
    "interaction_tone": "professional",
    "summary": "Portfolio site showcasing mobile app design work",
    "discovery_method": "progressive_enhancement"
  }
}

Strategy: Start with standard MCP, add web discovery layer, enhance progressively.

Business Website

{
  "feed_type": "mcp", 
  "metadata": {
    "title": "Bakery Corner",
    "origin": "https://bakerycorner.com",
    "description": "Local bakery in downtown Portland"
  },
  "prompts": [
    {
      "intent": "get_hours",
      "keywords": ["hours", "open", "when", "schedule"],
      "description": "Show bakery opening hours and contact info"
    }
  ],
  "agent_guidance": {
    "fallback_behavior": "suggest calling for current availability"
  }
}

Documentation Site

{
  "feed_type": "mcp",
  "metadata": {
    "title": "DevTools Documentation", 
    "origin": "https://docs.devtools.com",
    "description": "API documentation and guides for developers"
  },
  "prompts": [
    {
      "intent": "search_docs",
      "keywords": ["how to", "guide", "tutorial", "API"],
      "description": "Help users find relevant documentation"
    }
  ],
  "capabilities": [
    {
      "name": "searchDocs",
      "method": "GET",
      "path": "/api/search",
      "description": "Search documentation content"
    }
  ]
}

Trust & Signatures

Basic Trust Declaration

{
  "feed_type": "mcp",
  "metadata": {
    "title": "My Blog",
    "origin": "https://myblog.com"
  },
  "trust": {
    "signed_blocks": ["metadata", "trust"],
    "scope": "public"
  },
  "signature": {
    "value": "signature_hash_here",
    "created_at": "2025-06-10T14:30:00Z"
  }
}

Certified by LLMCA

{
  "trust": {
    "signed_blocks": ["metadata", "prompts", "trust"],
    "scope": "public",
    "certifier": "https://llmca.org"
  },
  "certification": {
    "issuer": "https://llmca.org",
    "cert_id": "llmca-2025-001",
    "issued_at": "2025-06-10T10:00:00Z"
  }
}

Why sign your MCP feed?

• ✅ Prove authenticity to agents
• ✅ Prevent tampering
• ✅ Build trust reputation
• ✅ Enable verification by users

Core Fields Reference

Discovery by Agents *(Updated Reality)*

Current MCP practices (Anthropic):

• Local configuration files only
• No web discovery standard established
• stdin/stdout and JSON-RPC connections

LLMFeed innovation (Web-native):

1. https://yoursite.com/.well-known/mcp.llmfeed.json (primary)
2. https://yoursite.com/.well-known/llm-index.llmfeed.json (for feed discovery)

What agents do with LLMFeed:

1. Discover via .well-known/ (web-based) ✅
2. Parse enhanced MCP format ✅
3. Use behavioral guidance and trust signals ✅
4. Request permission for enhanced features (current) 🛡️
5. Enable autonomy for signed content (future) 🔮

Innovation positioning: LLMFeed bridges MCP's excellent tool calling with web-scale discovery and trust.

📚 Advanced Features

{
  "capabilities": [
    {
      "name": "searchProducts",
      "method": "GET", 
      "path": "/api/search",
      "description": "Search product catalog",
      "requires_user_consent": false
    },
    {
      "name": "submitOrder",
      "method": "POST",
      "path": "/api/orders",
      "description": "Submit a product order",
      "requires_user_consent": true
    }
  ]
}

{
  "agent_services": {
    "keywords": ["support", "help", "contact"],
    "action_endpoint": "https://example.org/api/contact",
    "user_info_fields": ["name", "email", "message"],
    "requires_consent": true
  }
}

{
  "intent_router": [
    {
      "intent": "learn_about_products",
      "triggers": [
        "what do you sell",
        "show me products", 
        "catalog"
      ],
      "response_hint": "Show product categories and highlight featured items",
      "suggested_targets": [
        {
          "title": "Product Catalog",
          "url": "/products"
        }
      ]
    },
    {
      "intent": "get_support", 
      "triggers": [
        "help",
        "support",
        "problem"
      ],
      "response_hint": "Offer to connect with support team",
      "suggested_targets": [
        {
          "title": "Contact Support",
          "url": "/contact"
        }
      ]
    }
  ]
}

{
  "booking_slots": {
    "endpoint": "https://mybusiness.com/api/slots",
    "available_hours": ["2025-06-01T10:00Z", "2025-06-01T14:00Z"],
    "duration_minutes": 30,
    "fields_required": ["name", "email", "service_type"]
  }
}

{
  "regional_presence": {
    "region": "us-west",
    "languages": ["en", "es"],
    "support": {
      "available": true,
      "contact": "mailto:support@mybusiness.com",
      "hours": "9AM-5PM PST"
    }
  }
}

{
  "capabilities": [
    {
      "type": "endpoint",
      "intent": "get status",
      "url": "https://api.example.com/status",
      "description": "Check service health"
    },
    {
      "type": "openapi",
      "url": "https://example.com/.well-known/openapi.json", 
      "description": "Complete API specification"
    }
  ]
}

{
  "feed_type": "mcp",
  "metadata": {
    "title": "Pizza Palace",
    "description": "Local pizza delivery"
  },
  "prompts": [
    {
      "intent": "order_pizza",
      "keywords": ["pizza", "order", "delivery"]
    }
  ]
}

Best Practices

Progressive Enhancement Strategy

1. Maintain MCP compatibility - existing tools keep working
2. Add web discovery layer - enhanced agents get additional features
3. Test with consent model - validate semi-automatic discovery
4. Prepare for trust evolution - plan signature implementation

For Current Deployment

1. Copy + extend approach - duplicate MCP content to .well-known/
2. Enhanced behavioral guidance - help agents interact appropriately
3. Signature readiness - prepare for future autonomous capabilities
4. Multi-agent compatibility - works beyond Claude ecosystem

For Business Sites

1. Declare your services clearly in metadata
2. Add contact/booking capabilities if relevant
3. Use regional_presence for local businesses
4. Consider certification for trust-critical services

For Developers

1. Separate complex APIs into capabilities.llmfeed.json
2. Link to OpenAPI specs for technical details
3. Test with multiple LLMs before deployment
4. Version your feeds for iterative improvement

Strategic Positioning

Anthropic MCP: Excellent foundation for tool calling
LLMFeed enhancement: Adds web discovery + trust + behavioral guidance
Result: Migration path from assisted AI to autonomous agents

Related Feed Types

Think of MCP as the "front door" that guides agents to other specialized feeds.

What This Enables

For Site Owners

• ✅ Control how agents interpret your site
• ✅ Guide agent behavior appropriately
• ✅ Build trust through signatures
• ✅ Attract agent-based traffic
• ✅ Future-ready infrastructure for autonomous agents

For Users

• ✅ Better agent responses about your site
• ✅ Clearer capabilities and limitations
• ✅ Trustworthy interactions through verification
• ✅ Appropriate fallbacks when agents can't help

For Agents

• ✅ Understand site purpose without guessing
• ✅ Respect owner intentions and limitations
• ✅ Verify information authenticity
• ✅ Provide better user experiences
• ✅ Enable safe autonomous operation (future)

References

• Official Anthropic MCP – Authoritative MCP specification
• Getting Started Guide
• Capabilities Feed Type
• Well-Known Discovery
• Trust & Signatures