💊 The Information Capsule Concept

Universal Application Sources

Export feeds transform any information into structured capsules that LLMs can understand:

Desktop App → Export Feed → LLM understands context
Database Query → Export Feed → Agent processes results  
User Conversation → Export Feed → Transfer to another agent
System Logs → Export Feed → AI analyzes patterns

What Makes a Good Information Capsule

1. Context: Not just data, but where it came from and why
2. Structure: Organized for machine processing
3. Metadata: Creation time, origin, purpose, tags
4. Trust Information: Signature status, verification hints
5. Usage Hints: How the LLM should interpret or use the data

Beyond Web Exports

While export buttons on websites are common, the format works for any application:

• Electron apps can export user projects as .llmfeed.json
• Mobile apps can share user data with proper consent
• CLI tools can output structured reports for agent analysis
• Desktop software can create portable context for AI assistants
• IoT devices can export sensor data with metadata

The key is packaging information with intent so any LLM receiving the capsule understands its purpose and provenance.

🔏 Signature and Trust

Why Sign Export Feeds?

Signatures provide three critical guarantees:

1. Ownership: Proves who created the export
2. Integrity: Ensures content hasn't been tampered with
3. Trust: Allows LLMs and agents to assess reliability

When to Sign

Trust Levels

{
  "trust": {
    "trust_level": "self-declared",     // or "certified"
    "scope": "partial",                 // or "complete"
    "signed_blocks": ["metadata", "content"],
    "certifier": "https://example.com/.well-known/public.pem"
  }
}

Note: Unsigned exports are still valid but may be treated with lower trust by security-conscious agents and enterprise systems.

📄 Basic metadata

title: "Feed Type: export.llmfeed.json"
description: "MCP documentation on Feed Type: export.llmfeed.json - Universal information capsules for LLM consumption"
date: "2025-06-17T10:00:00.000Z"
lang: "en"

🏷️ Tags and classification

tags:

• "mcp"
• "ai-agents"
• "security"
• "data-classification"
• "information-capsules"
  format: "documentation"
  category: "technical"
  contentType: "documentation"

🧠 Intent and audience

intent: "inform"
llmIntent: "browse-spec"
llmTopic: "specification"
audience:

• "llm"
• "developer"
• "security-engineer"
• "enterprise-architect"

📊 Page properties

pageType: "documentation"
interactionComplexity: "simple"
priority: "normal"
riskLevel: "low"
updateFrequency: "static"

🔗 URLs

slug: "llmfeed_feedtype_export"
canonical_url: "https://wellknownmcp.org/spec/02_llmfeed_feedtype/llmfeed_feedtype_export "
githubUrl: "https://github.com/wellknownmcp/llmfeed-spec/blob/main/02_llmfeed_feedtype/llmfeed_feedtype_export.md "
mcpFeedUrl: "/.well-known/mcp.llmfeed.json"

🤖 Agent optimization

autoDiscoverFeeds: true
agentReadiness: true
llmBehaviorHints: "suggest-only"

📋 Capabilities

capabilities: - "verification" - "export" - "feed-generation" - "security-classification" - "information-capsules"

Feed Type: `export.llmfeed.json`

Purpose

This feed creates information capsules from any application or data source — ready for LLM consumption, agent processing, or secure transfer.

Export feeds can originate from:

• Web applications: dashboards, documentation, user interfaces
• Desktop software: documents, databases, project files
• Mobile apps: user data, settings, conversations
• Command-line tools: logs, reports, system information
• APIs and services: structured data, responses, metadata

The core concept is packaging information with context so LLMs can understand not just the data, but its origin, purpose, and trustworthiness.

Security Extension: Supports automatic data classification and secure export workflows for enterprise environments.

Location

Typical path:

/exports/faq.llmfeed.json

Can be linked from:

• llm-index.llmfeed.json
• buttons on site (ExportToLLM)
• internal agent menus

Basic Structure

{
  "feed_type": "export",
  "metadata": {
    "title": "FAQ",
    "origin": "https://example.org",
    "description": "Frequently asked questions",
    "generated_at": "2025-06-17T10:30:00Z"
  },
  "summary": "This FAQ explains the trust system and how to verify signed feeds.",
  "tags": ["faq", "documentation", "trust"],
  "trust": { ... }
}

🔐 Security-Enhanced Export Structure

For exports containing potentially sensitive data, the format extends with security metadata:

{
  "feed_type": "export",
  "metadata": {
    "title": "Page Export with Security",
    "origin": "https://example.com/secure-page",
    "generated_at": "2025-06-17T10:30:00Z"
  },
  "data_classification": {
    "security_scan_performed": true,
    "sensitive_data_handling": "user_consented",
    "redacted_fields": ["api_keys", "internal_urls"],
    "warning_shown": true,
    "user_consent": {
      "timestamp": "2025-06-17T10:29:45Z",
      "items_approved": ["email_addresses", "user_preferences"],
      "items_rejected": ["internal_system_ids"]
    }
  },
  "content": {
    "processed_html": "...", 
    "metadata_extracted": "...",
    "sensitive_placeholders": {
      "[API_KEY_REDACTED]": "Original contained an API key",
      "[INTERNAL_URL_REDACTED]": "Internal URL masked for security"
    }
  },
  "trust": {
    "data_integrity": "verified",
    "sanitization_performed": true,
    "risk_level": "low"
  }
}

Data Classification Levels

Security Workflow

1. Automatic Scan: Content is scanned for sensitive patterns
2. Classification: Data is categorized by sensitivity level
3. User Consent: For sensitive data, user chooses what to include
4. Secure Export: Generate feed with appropriate redaction and metadata

Modes of Generation

Universal Application: These modes work for any type of application — web, desktop, mobile, or command-line. The export mechanism adapts to the platform while maintaining the same feed structure.

⚠️ Security Note: Signature is recommended for static exports, and required for dynamic exports containing sensitive data.

🧳 Structured Bundles (`data.files[]`)

An export feed may describe the contents of an archive (ZIP) via a data.files[] block.

Minimal example (structure only):

{
  "feed_type": "export",
  "metadata": { "title": "Bundle Index" },
  "data": {
    "files": [
      { "path": "README.md" },
      { "path": "src/index.js" },
      { "path": "images/logo.png" }
    ]
  }
}

Security-enhanced bundle:

{
  "data": {
    "files": [
      {
        "path": "src/config.js",
        "tags": ["code", "configuration"],
        "description": "Application configuration",
        "security_classification": "sensitive",
        "redaction_applied": "credentials_masked"
      },
      {
        "path": "README.md",
        "tags": ["documentation", "public"],
        "description": "Project documentation",
        "security_classification": "public"
      }
    ],
    "security_summary": {
      "total_files": 2,
      "public_files": 1,
      "sensitive_files": 1,
      "critical_files": 0
    }
  }
}

🎯 Export Use Cases

Simple Documentation Export

{
  "feed_type": "export",
  "metadata": {
    "title": "API Documentation",
    "origin": "https://api.example.com/docs"
  },
  "content": {
    "documentation": "Complete API reference...",
    "endpoints": [...],
    "examples": [...]
  }
}

Secure Credential Export

{
  "feed_type": "export",
  "metadata": {
    "title": "API Access Package",
    "origin": "https://dashboard.example.com"
  },
  "data_classification": {
    "security_scan_performed": true,
    "sensitive_data_handling": "admin_approved",
    "classification_level": "restricted"
  },
  "content": {
    "api_endpoint": "https://api.example.com",
    "key_hint": "sk_live_abc***",
    "permissions": ["read", "write"],
    "rate_limits": "1000/hour"
  },
  "trust": {
    "signed_blocks": ["content", "metadata"],
    "certifier": "https://example.com/.well-known/public.pem"
  }
}

Page Context Export with Privacy

{
  "feed_type": "export",
  "metadata": {
    "title": "Dashboard Export",
    "origin": "https://app.example.com/dashboard"
  },
  "data_classification": {
    "security_scan_performed": true,
    "sensitive_data_handling": "user_consented",
    "user_consent": {
      "email_addresses": true,
      "user_preferences": true,
      "internal_ids": false
    }
  },
  "content": {
    "dashboard_data": "User preferences and settings...",
    "user_email": "user@example.com",
    "internal_user_id": "[USER_ID_REDACTED]"
  }
}

🛡️ Enterprise Security Features

Automatic Pattern Detection

const CRITICAL_PATTERNS = {
  api_keys: /^(sk_|pk_|key_|token_|secret_)/i,
  passwords: /password|pwd|pass/i,
  private_keys: /-----BEGIN.*PRIVATE KEY-----/,
  session_tokens: /sess_|session_/i
};

const SENSITIVE_PATTERNS = {
  emails: /\b[A-Za-z0-9._%+-]+@[A-Za-z0-9.-]+\.[A-Z|a-z]{2,}\b/g,
  phone_numbers: /\b\d{3}[-.]?\d{3}[-.]?\d{4}\b/g,
  internal_urls: /https?:\/\/internal\.|localhost/g
};

Compliance Integration

• GDPR: Automatic consent management for PII
• SOX: Audit trail for financial data exports
• HIPAA: PHI detection and handling
• Corporate Policies: Custom pattern detection

Best Practices

Security

• ✅ Always scan for sensitive data before export
• ✅ Implement user consent for personal information
• ✅ Use placeholders for redacted content
• ✅ Sign exports containing any sensitive data
• ✅ Log all export activities for audit

Performance

• ✅ Use tags to describe content type
• ✅ Inline small content, reference large files
• ✅ Include file metadata for bundles
• ✅ Reference via llm-index for discoverability

User Experience

• ✅ Clear consent interfaces for sensitive data
• ✅ Preview what will be exported
• ✅ Explain why data is being redacted
• ✅ Provide export without sensitive data option

🚀 Future: Progressive Integration Levels

Export feeds are the foundation of a progressive integration strategy:

• Level 1-2: Inline + file exports with security (✅ implemented)
• Level 3: Universal export buttons with consent UX
• Level 4-5: Browser & OS integration (See Vision →)

For enterprise security considerations, see our Enterprise Roadmap.

Related

• `llmfeed.md`
• `https://wellknownmcp.org/tools/export-button`
• `llmfeed_feedtype_credential.md`
• `07_vision/enterprise-security-roadmap.md`