MCP Integration

The AI Provenance Protocol (APP) and the Model Context Protocol (MCP) are complementary standards. MCP defines how AI applications connect to external tools and data sources; APP records the provenance of AI-generated output.

How they fit together

Concern	MCP	APP
AI-to-tool connectivity	Primary purpose	Not covered
Provenance of AI output	Not covered	Primary purpose
Content verification	Not covered	Verification protocol
Human review tracking	Not covered	Review object
EU AI Act compliance	Not directly addressed	Core design target

MCP servers that generate content

If you build an MCP server that uses AI models to generate content, include APP metadata in your tool responses.

Example: Tool response with APP metadata


{
  "content": [
    {
      "type": "text",
      "text": "Generated product description..."
    }
  ],
  "_ai_provenance": {
    "app_version": "1.0.0",
    "ai_generated": true,
    "generator": {
      "platform": "product-description-server",
      "model": "anthropic/claude-sonnet-4"
    },
    "generated_at": "2026-03-06T14:22:00Z",
    "generation_id": "a1b2c3d4-e5f6-7890-abcd-ef1234567890"
  }
}

MCP extension namespace

Store MCP-specific context in the io.modelcontextprotocol extension namespace:


{
  "extensions": {
    "io.modelcontextprotocol": {
      "server_name": "product-description-server",
      "server_version": "1.0.0",
      "tool_name": "generate_description",
      "session_id": "mcp-session-abc123"
    }
  }
}

Provenance chains

When MCP workflows involve multiple AI generation steps, use parent_generation_id to create provenance chains:


Step 1: "research" tool      → generation_id: "aaa-111"
Step 2: "write_draft" tool    → generation_id: "bbb-222", parent: "aaa-111"
Step 3: "polish" tool         → generation_id: "ccc-333", parent: "bbb-222"

Best practices

Always include APP metadata when your MCP server uses AI models
Use the MCP extension namespace for MCP-specific context
Create provenance chains when multiple tools contribute to output
Expose verification endpoints for regulated contexts
Record human review when MCP tool output is reviewed before use