Audit Log Schema

Golf Gateway generates structured audit logs for all MCP traffic, authentication events, and security analysis. This reference documents the complete schema used for audit log entries.

Quick Reference

Category	Key Fields	Purpose
Core	`timestamp`, `event_type`, `component`	Event identification and timing
Identity	`session_id`, `gateway_id`, `organization_id`	Session and gateway tracking
User	`user.sub`, `user.email`, `user.authenticated`	User identity (privacy-compliant)
Authentication	`authentication.result`, `authentication.provider_type`	Auth method and result
Security	`security.threat_score`, `security.blocked`, `security.was_scrubbed`	Threat detection and PII scrubbing
MCP Protocol	`mcp_method`, `mcp_tool_name`, `mcp_*_payload`	MCP JSON-RPC details
Integrity	`sequence`, `prev_hash`, `integrity_hash`	Tamper-proof hash chain
Internal	`message`, `correlation`, `metadata`	Internal processing (not exported)

Schema Overview

View Complete Schema Example

{
  "timestamp": "2024-01-15T10:30:45.123456+00:00",
  "event_type": "mcp_request",
  "component": "mcp_parser",
  "session_id": "sess_abc123def456",
  "gateway_id": "550e8400-e29b-41d4-a716-446655440000",
  "gateway_name": "production-gateway",
  "organization_id": "org_xyz789",
  "client_info": {
    "name": "claude-desktop",
    "version": "1.2.3"
  },
  "authentication": {
    "method": "oauth2",
    "result": "success",
    "completed_at": 1705315845123,
    "provider_type": "auth0",
    "provider_name": "Production Auth0",
    "authorization_result": "granted",
    "scopes": ["read:mcp", "write:mcp"],
    "granted_at": 1705315845123,
    "user_id": "auth0|abc123",
    "email": "a1b2c3d4e5f6g7h8",
    "session_created": true,
    "session_type": "authenticated"
  },
  "security": {
    "threat_score": 0.02,
    "threats_detected": [],
    "analyzed": true,
    "blocked": false,
    "detection_method": "llm",
    "content_size_bytes": 256,
    "inference_ms": 38.5,
    "was_scrubbed": false,
    "entities_redacted": 0,
    "fields_masked": 0
  },
  "request": {
    "method": "POST",
    "path": "/finance-tools/mcp",
    "size_bytes": 256,
    "headers": {}
  },
  "upstream": {
    "host": "mcp-server.internal",
    "port": 3001,
    "url": "http://mcp-server.internal:3001",
    "name": "finance-tools"
  },
  "user": {
    "sub": "auth0|abc123",
    "email": "a1b2c3d4e5f6g7h8",
    "authenticated": true,
    "anonymous": false
  },
  "mcp_method": "tools/call",
  "mcp_tool_name": "get_account_balance",
  "mcp_request_payload": "{\n  \"jsonrpc\": \"2.0\",\n  \"method\": \"tools/call\",\n  \"params\": {\n    \"name\": \"get_account_balance\",\n    \"arguments\": {\"account_id\": \"12345\"}\n  },\n  \"id\": 1\n}",
  "mcp_request_tokens": 45,
  "has_error": false,
  "sequence": 12345,
  "prev_hash": "a1b2c3d4e5f6a1b2c3d4e5f6a1b2c3d4e5f6a1b2c3d4e5f6a1b2c3d4e5f6a1b2",
  "integrity_hash": "d4e5f6a1b2c3d4e5f6a1b2c3d4e5f6a1b2c3d4e5f6a1b2c3d4e5f6a1b2c3d4e5"
}

Each audit log entry is a JSON object. The sections below document each field category in detail.

{
  "timestamp": "2024-01-15T10:30:45.123456+00:00",
  "event_type": "mcp_request",
  "component": "mcp_parser",
  "session_id": "sess_abc123",
  "gateway_id": "550e8400-e29b-41d4-a716-446655440000",
  "gateway_name": "production-gateway",
  "organization_id": "org_xyz789",
  "client_info": { ... },
  "authentication": { ... },
  "security": { ... },
  "request": { ... },
  "upstream": { ... },
  "user": { ... },
  "mcp_method": "tools/call",
  "mcp_tool_name": "get_weather",
  "mcp_request_payload": "{ ... }",
  "mcp_request_tokens": 150,
  "sequence": 12345,
  "prev_hash": "a1b2c3...",
  "integrity_hash": "d4e5f6..."
}

Core Fields

These fields are present on every audit log entry.

Field	Type	Description
`timestamp`	string	ISO 8601 timestamp with timezone (e.g., `2024-01-15T10:30:45.123456+00:00`)
`event_type`	string	Type of audit event (see Event Types)
`component`	string	System component that generated the event

Event Types

Value	Description
`mcp_request`	Standard MCP JSON-RPC request
`mcp_response`	Standard MCP JSON-RPC response
`mcp_sampling_request`	Sampling/createMessage request
`mcp_sampling_response`	Sampling/createMessage response
`mcp_elicitation_request`	Server-initiated elicitation request
`mcp_elicitation_response`	Elicitation response from client
`mcp_notification`	MCP notification (no response expected)
`authentication`	Authentication event
`security`	Security analysis event

Component Values

Value	Description
`mcp_parser`	MCP message parsing
`auth`	Authentication and authorization
`security`	Security analysis and threat detection

Session and Identity

Field	Type	Description
`session_id`	string \| null	MCP protocol session identifier
`gateway_id`	string \| null	Gateway UUID (multi-gateway deployments)
`gateway_name`	string \| null	Human-readable gateway name
`organization_id`	string \| null	Organization UUID (multi-tenant deployments)

Integrity Fields

These fields provide tamper-proof audit logging through cryptographic hash chaining.

Field	Type	Description
`sequence`	integer \| null	Monotonically increasing sequence number
`prev_hash`	string \| null	HMAC-SHA256 hash of the previous log entry
`integrity_hash`	string \| null	HMAC-SHA256 hash of the current entry

The integrity chain allows verification that no log entries have been modified or deleted. Configure the hash key with GOLF_SECURITY_INTEGRITY_KEY.

Client Info

Information about the MCP client, extracted from the initialize request.

Field	Type	Description
`client_info.name`	string	MCP client name (e.g., “claude-desktop”, “cursor”)
`client_info.version`	string	MCP client version

Example:

{
  "client_info": {
    "name": "claude-desktop",
    "version": "1.2.3"
  }
}

Authentication

Authentication and authorization results.

Field	Type	Description
`authentication.method`	string	Authentication method (default: `oauth2`)
`authentication.result`	string	`success`, `failed`, or `anonymous`
`authentication.completed_at`	integer	Unix timestamp in milliseconds
`authentication.provider_type`	string \| null	IDP type: `auth0`, `okta`, `microsoft-entra-id`, `descope`
`authentication.provider_name`	string \| null	Human-readable provider name
`authentication.authorization_result`	string	`granted`, `denied`, or `anonymous`
`authentication.scopes`	array	Granted OAuth scopes
`authentication.granted_at`	integer	Unix timestamp in milliseconds
`authentication.user_id`	string \| null	User subject identifier
`authentication.email`	string \| null	User email (hashed before export)
`authentication.name`	string \| null	User display name (removed before export)
`authentication.session_created`	boolean	Whether a new session was created
`authentication.session_type`	string	`authenticated` or `anonymous`

Example:

{
  "authentication": {
    "method": "oauth2",
    "result": "success",
    "completed_at": 1705315845123,
    "provider_type": "auth0",
    "provider_name": "Production Auth0",
    "authorization_result": "granted",
    "scopes": ["read:mcp", "write:mcp"],
    "granted_at": 1705315845123,
    "user_id": "auth0|abc123",
    "email": "a1b2c3d4e5f6g7h8",
    "session_created": true,
    "session_type": "authenticated"
  }
}

Security

Security analysis and threat detection results.

Field	Type	Description
`security.threat_score`	float	Threat score from 0.0 (safe) to 1.0 (malicious)
`security.threats_detected`	array	List of detected threat types
`security.analyzed`	boolean	Whether security analysis was performed
`security.skipped_reason`	string \| null	Reason analysis was skipped
`security.blocked`	boolean	Whether the request/response was blocked
`security.detection_method`	string \| null	`llm`, `pattern`, or `rate_limit`
`security.threat_category`	string \| null	`prompt_injection`, `jailbreak`, `data_exfiltration`, etc.
`security.content_size_bytes`	integer \| null	Size of analyzed content in bytes
`security.inference_ms`	float \| null	LLM inference duration in milliseconds

PII Scrubbing Fields

Field	Type	Description
`security.was_scrubbed`	boolean	Whether content was scrubbed for PII
`security.entities_redacted`	integer	Total PII entities redacted
`security.fields_masked`	integer	Total fields masked by name
`security.entities_by_type`	object	Breakdown of redacted entities by type
`security.redacted_paths`	array	JSON paths where entities were redacted
`security.masked_field_names`	array	Field names that were masked
`security.custom_rules_fired`	array	Custom regex rules that matched
`security.scrub_duration_ms`	float \| null	Scrubbing duration in milliseconds

Example (threat detected):

{
  "security": {
    "threat_score": 0.95,
    "threats_detected": ["prompt_injection"],
    "analyzed": true,
    "blocked": true,
    "detection_method": "llm",
    "threat_category": "prompt_injection",
    "content_size_bytes": 1024,
    "inference_ms": 45.2
  }
}

Example (PII scrubbed):

{
  "security": {
    "threat_score": 0.0,
    "threats_detected": [],
    "analyzed": true,
    "blocked": false,
    "was_scrubbed": true,
    "entities_redacted": 3,
    "fields_masked": 1,
    "entities_by_type": {
      "email": 2,
      "phone": 1
    },
    "redacted_paths": ["result.content[0].text"],
    "masked_field_names": ["ssn"],
    "scrub_duration_ms": 12.5
  }
}

Request

Request context information.

Field	Type	Description
`request.method`	string \| null	HTTP or MCP method
`request.path`	string \| null	Request path or resource
`request.size_bytes`	integer \| null	Request size in bytes
`request.headers`	object	Request headers (sanitized to empty before export)

Upstream

Information about the upstream MCP server.

Field	Type	Description
`upstream.host`	string	Upstream server hostname
`upstream.port`	integer	Upstream server port (1-65535)
`upstream.url`	string \| null	Original URL for display
`upstream.name`	string \| null	Upstream server name

Example:

{
  "upstream": {
    "host": "mcp-server.internal",
    "port": 3001,
    "name": "finance-tools"
  }
}

User

User identity information.

Field	Type	Description
`user.sub`	string \| null	User subject identifier
`user.email`	string \| null	User email (hashed to 16-char SHA256 before export)
`user.name`	string \| null	User display name (removed before export)
`user.authenticated`	boolean	Whether user is authenticated
`user.anonymous`	boolean	Whether user is anonymous

For privacy compliance, user.email is hashed and user.name is removed before export to external systems.

MCP Protocol Fields

Extracted Fields

These fields are extracted from MCP messages for efficient querying and aggregation.

Field	Type	Description
`mcp_method`	string \| null	MCP JSON-RPC method (e.g., `tools/call`, `resources/read`)
`mcp_tool_name`	string \| null	Tool name from `tools/call` or `tools/list`
`mcp_resource_uri`	string \| null	Resource URI from `resources/read` or `resources/list`
`mcp_prompt_name`	string \| null	Prompt name from `prompts/get` or `prompts/list`

Raw Payloads

Field	Type	Description
`mcp_request_payload`	string \| null	Raw MCP request JSON (formatted)
`mcp_response_payload`	string \| null	Raw MCP response JSON (formatted)

Token Estimates

Field	Type	Description
`mcp_request_tokens`	integer \| null	Estimated token count for request
`mcp_response_tokens`	integer \| null	Estimated token count for response

Error Detection

Field	Type	Description
`has_error`	boolean \| null	Whether MCP response contains an error

Elicitation Fields

Fields for server-initiated user input requests (MCP elicitation).

Field	Type	Description
`mcp_elicitation_mode`	string \| null	Elicitation mode: `form` or `url`
`mcp_elicitation_id`	string \| null	Elicitation ID for URL mode correlation
`mcp_elicitation_action`	string \| null	User action: `accept`, `decline`, or `cancel`

Threat Analysis

Detailed threat analysis results (when available).

Field	Type	Description
`threat_analysis.threat_score`	float	Threat score from 0.0 to 1.0
`threat_analysis.detected_patterns`	array	List of detected threat patterns
`threat_analysis.severity`	string	`critical`, `incident`, or `low`
`threat_analysis.remediation`	string \| null	Suggested remediation action

Session Statistics

Aggregate session statistics (for monitoring events).

Field	Type	Description
`session_stats.active_sessions`	integer	Number of active sessions
`session_stats.total_created`	integer	Total sessions created
`session_stats.total_authenticated`	integer	Total authenticated sessions
`session_stats.total_anonymous`	integer	Total anonymous sessions

Internal Fields

These fields are present in audit log entries for internal processing but are excluded from external exports and the GraphQL API.

Field	Type	Description
`message`	object \| null	Raw MCP message data for internal processing
`correlation`	object \| null	Request/response correlation for internal tracking
`metadata`	object	Additional metadata for legacy compatibility and flexible extensions

These fields are for internal use only. They are not exposed through the GraphQL API and are excluded from export payloads to external systems (Elasticsearch, OpenTelemetry, Sentinel).

Complete Example

MCP Request Event:

{
  "timestamp": "2024-01-15T10:30:45.123456+00:00",
  "event_type": "mcp_request",
  "component": "mcp_parser",
  "session_id": "sess_abc123def456",
  "gateway_id": "550e8400-e29b-41d4-a716-446655440000",
  "gateway_name": "production-gateway",
  "organization_id": "org_xyz789",
  "client_info": {
    "name": "claude-desktop",
    "version": "1.2.3"
  },
  "authentication": {
    "method": "oauth2",
    "result": "success",
    "completed_at": 1705315845123,
    "provider_type": "auth0",
    "authorization_result": "granted",
    "scopes": ["read:mcp", "write:mcp"],
    "granted_at": 1705315845123,
    "user_id": "auth0|abc123",
    "session_type": "authenticated"
  },
  "security": {
    "threat_score": 0.02,
    "threats_detected": [],
    "analyzed": true,
    "blocked": false,
    "detection_method": "llm",
    "content_size_bytes": 256,
    "inference_ms": 38.5
  },
  "upstream": {
    "host": "mcp-server.internal",
    "port": 3001,
    "name": "finance-tools"
  },
  "user": {
    "sub": "auth0|abc123",
    "authenticated": true,
    "anonymous": false
  },
  "mcp_method": "tools/call",
  "mcp_tool_name": "get_account_balance",
  "mcp_request_payload": "{\n  \"jsonrpc\": \"2.0\",\n  \"method\": \"tools/call\",\n  \"params\": {\n    \"name\": \"get_account_balance\",\n    \"arguments\": {\"account_id\": \"12345\"}\n  },\n  \"id\": 1\n}",
  "mcp_request_tokens": 45,
  "has_error": false,
  "sequence": 12345,
  "prev_hash": "a1b2c3d4e5f6a1b2c3d4e5f6a1b2c3d4e5f6a1b2c3d4e5f6a1b2c3d4e5f6a1b2",
  "integrity_hash": "d4e5f6a1b2c3d4e5f6a1b2c3d4e5f6a1b2c3d4e5f6a1b2c3d4e5f6a1b2c3d4e5"
}

Security Blocked Event:

{
  "timestamp": "2024-01-15T10:31:12.789012+00:00",
  "event_type": "mcp_request",
  "component": "security",
  "session_id": "sess_xyz789abc123",
  "gateway_name": "production-gateway",
  "security": {
    "threat_score": 0.97,
    "threats_detected": ["prompt_injection"],
    "analyzed": true,
    "blocked": true,
    "detection_method": "llm",
    "threat_category": "prompt_injection",
    "content_size_bytes": 2048,
    "inference_ms": 52.3
  },
  "threat_analysis": {
    "threat_score": 0.97,
    "detected_patterns": ["instruction_override", "role_manipulation"],
    "severity": "critical",
    "remediation": "Block request and notify security team"
  },
  "mcp_method": "tools/call",
  "mcp_tool_name": "execute_query",
  "sequence": 12346,
  "integrity_hash": "f6a1b2c3d4e5f6a1b2c3d4e5f6a1b2c3d4e5f6a1b2c3d4e5f6a1b2c3d4e5f6a1"
}

Export Destinations

Audit logs are exported to configured destinations with automatic schema mapping:

Destination	Index Pattern	Documentation
Elasticsearch	`{prefix}-YYYY.MM.DD`	Export to Elasticsearch
OpenTelemetry	OTLP logs	Export to OpenTelemetry
Microsoft Sentinel	Custom DCR table	Export to Sentinel

Each destination applies field-specific transformations. For example, Elasticsearch uses daily indices while Sentinel transforms fields to PascalCase for KQL compatibility.

Audit Log Schema

Audit Log Schema

Quick Reference

Schema Overview

Core Fields

Event Types

Component Values

Session and Identity

Integrity Fields

Client Info

Authentication

Security

PII Scrubbing Fields

Request

Upstream

User

MCP Protocol Fields

Extracted Fields

Raw Payloads

Token Estimates

Error Detection

Elicitation Fields

Threat Analysis

Session Statistics

Internal Fields

Complete Example

Export Destinations

Environment Variables

YAML Configuration

Export to Elasticsearch

Export to Sentinel

​Audit Log Schema

​Quick Reference

​Schema Overview

​Core Fields

​Event Types

​Component Values

​Session and Identity

​Integrity Fields

​Client Info

​Authentication

​Security

​PII Scrubbing Fields

​Request

​Upstream

​User

​MCP Protocol Fields

​Extracted Fields

​Raw Payloads

​Token Estimates

​Error Detection

​Elicitation Fields

​Threat Analysis

​Session Statistics

​Internal Fields

​Complete Example

​Export Destinations

​Related Documentation

Environment Variables

YAML Configuration

Export to Elasticsearch

Export to Sentinel

Audit Log Schema

Quick Reference

Schema Overview

Core Fields

Event Types

Component Values

Session and Identity

Integrity Fields

Client Info

Authentication

Security

PII Scrubbing Fields

Request

Upstream

User

MCP Protocol Fields

Extracted Fields

Raw Payloads

Token Estimates

Error Detection

Elicitation Fields

Threat Analysis

Session Statistics

Internal Fields

Complete Example

Export Destinations

Related Documentation