Skip to content

Agent Proxy generates workload events for the following protocols and technologies:

  • HTTP
  • Snowflake
  • MCP
  • PostgreSQL
  • Amazon Redshift
  • MySQL and MariaDB
  • Oracle Database
  • Redis
  • TCP Passthrough

This page lists the application fields Agent Proxy records for each protocol, split by request and response event. For the common fields that every event shares, see the Workload Event reference. For a conceptual overview, see Workload Events.

The Presence column indicates whether Agent Proxy always includes the field (Always) or includes it only under certain conditions (Optional).

HTTP fields appear under application.http. Snowflake events are HTTP-based queries to Snowflake endpoints and use the same fields, under application.snowflake.

Request fields:

FieldPresenceDescription
httpMethodAlwaysHTTP method, such as GET or POST.
httpVersionAlwaysHTTP version, such as HTTP/1.1.
uriAlwaysRequested URI, with the query string removed to avoid recording sensitive values.
headersAlwaysA list of single-key objects, each a header name and value, with sensitive headers redacted.

Response fields:

FieldPresenceDescription
httpVersionAlwaysHTTP version, such as HTTP/1.1.
headersAlwaysA list of single-key objects, each a header name and value, with sensitive headers redacted.
httpResponseCodeAlwaysHTTP response status code, such as 200 or 404.
Example application.http block
"application": {
"http": {
"httpMethod": "GET",
"httpVersion": "HTTP/1.1",
"uri": "/v2/accounts",
"headers": [
{"content-type": "application/json"},
{"host": "api.example.com"}
],
"httpResponseCode": 200
}
}

MCP events use the HTTP-based fields under application.mcp, plus MCP-specific fields.

Request fields:

FieldPresenceDescription
httpMethodAlwaysHTTP method, such as POST.
httpVersionAlwaysHTTP version, such as HTTP/1.1.
uriAlwaysRequested URI, with the query string removed to avoid recording sensitive values.
headersAlwaysA list of single-key objects, each a header name and value, with sensitive headers redacted.
mcpMethodAlwaysThe MCP method for the request, such as tools/list or tools/call.
mcpSessionIdAlwaysThe MCP session ID, shared across requests and responses in the same session.
mcpProtocolVersionOptionalThe negotiated MCP protocol version, when available.
mcpServerNameOptionalThe name of the target MCP server, when available.
mcpToolNameOptionalThe tool invoked, on tools/call requests.
mcpRequestIdOptionalThe MCP request identifier, used to correlate a response with its request.
userIdOptionalThe user identity for MCP flows that involve a human identity, matching the User (MCP App Protocol only) filter.

Response fields:

FieldPresenceDescription
httpVersionAlwaysHTTP version, such as HTTP/1.1.
headersAlwaysA list of single-key objects, each a header name and value, with sensitive headers redacted.
httpResponseCodeAlwaysHTTP response status code, such as 200 or 404.
mcpMethodAlwaysThe MCP method the response corresponds to, such as tools/list or tools/call.
mcpSessionIdAlwaysThe MCP session ID, shared across requests and responses in the same session.
mcpProtocolVersionOptionalThe negotiated MCP protocol version, when available.
mcpServerNameOptionalThe name of the target MCP server, when available.
mcpRequestIdOptionalThe MCP request identifier, used to correlate the response with its request.
Example application.mcp block
"application": {
"mcp": {
"httpMethod": "POST",
"httpVersion": "HTTP/1.1",
"uri": "/mcp/v1/tools/call",
"headers": [
{"content-type": "application/json"},
{"host": "mcp.example.com"}
],
"mcpMethod": "tools/call",
"mcpSessionId": "3d9c1e77-5b42-4c8a-9f16-2a7e8b0d4c31",
"mcpProtocolVersion": "2025-06-18",
"mcpServerName": "example-mcp-server",
"mcpToolName": "search_records",
"mcpRequestId": "42",
"userId": "alice@example.com"
}
}

When the Access Policy that governs MCP Identity Gateway: A component that brokers MCP traffic between MCP clients and target MCP servers, validating authorization and presenting Aembit-managed credentials on each request.Learn more traffic includes a Content Security: Content Security is an Access Policy component that inspects content and enforces the allow, block, or transform verdict returned by an inspection service.Learn more component, MCP workload events record the inspection decision under application.mcp.contentSecurity. For the CrowdStrike AIDR: CrowdStrike AI Detection and Response (AIDR) is a content inspection service that scans MCP tool listings, tool call inputs, and tool call outputs in real time and returns an allow, block, or transform verdict to Aembit.Learn more integration, the crowdStrikeAidr object contains these fields:

  • decision - The decision CrowdStrike AIDR returned for the inspected content.
  • requestId - CrowdStrike’s identifier for the inspection request. Use it to correlate the event with the corresponding finding in the CrowdStrike console.
  • reason - The reason for a blocked decision (for example, Blocked by AIDR policy).

The event’s outcome.result describes how the Gateway handled the message overall, so it doesn’t map one-to-one to the AIDR decision:

  • Passthrough with decision: allowed - AIDR allowed a response (a tool listing or tool result), and the Gateway relayed it to the client unchanged.
  • Modified with decision: allowed - AIDR allowed a request, but the Gateway still rewrote it to inject credentials toward the MCP server. Because credential injection applies to most proxied requests, allowed requests typically show Modified, not Passthrough.
  • Modified with decision: transformed - AIDR redacted the content, and the Gateway forwarded the transformed version.
  • Error with decision: blocked - AIDR blocked the content, and the Gateway sent the client an error response in its place. The crowdStrikeAidr.reason field carries the block reason.
  • Error with no decision - CrowdStrike AIDR returned an error or was unreachable, and the component blocked the content (fail closed). The event’s reason field carries the error detail.

Workload events record metadata about the decision, never the MCP request or response content that AIDR evaluated.

Example mcp.response event with a transformed tool result
{
"version": "1.0.0",
"meta": {
"timestamp": "2026-03-15T14:22:08.123Z",
"eventType": "mcp.response",
"connectionId": "3f8b2e1-7c4d-4a91-b5e6-9d2f1c8a4b73",
"eventId": "b2c8d4e6-3a1f-4b7c-9d5e-8f2a6c4b1d73",
"policyId": "dd987f8c-34fb-43e2-9d43-89d862e6b7ec",
"action": "forward",
"severity": "Info",
"resourceSetId": "ffffffff-ffff-ffff-ffff-ffffffffffff"
},
"workload": {
"client": { "id": "7c466803-9dd4-4388-9e45-420c57a0432c", "name": "MCP Identity Gateway" },
"server": { "id": "49183921-55ab-4856-a8fc-a032af695e0d", "name": "MCP Remote Server" }
},
"outcome": {
"result": "Modified"
},
"application": {
"mcp": {
"contentSecurity": {
"crowdStrikeAidr": {
"decision": "transformed",
"requestId": "prq_example1234567890abcdefghijklmn"
}
},
"httpResponseCode": 200,
"httpVersion": "HTTP/1.1",
"mcpMethod": "tools/call",
"mcpProtocolVersion": "2025-11-25",
"mcpRequestId": 1,
"mcpServerName": "MCP Remote Server",
"mcpSessionId": "eec9c934-55b4-4797-9993-36dfa4838399",
"mcpToolName": "example_tool",
"userId": "user@example.com"
}
},
"authorizationChain": [
"7fcd3dcc-dc4a-4e35-94e0-76221e12b83a",
"64d74e7a-19aa-4231-8286-5687b8959ecd",
"4b8be4ae-aa16-4003-965b-18fd7fab390b"
]
}

Correlate events with CrowdStrike AIDR findings

Section titled “Correlate events with CrowdStrike AIDR findings”

CrowdStrike AIDR keeps its own findings for the content it inspects. The requestId field links an Aembit workload event to the matching CrowdStrike record:

  1. In your Aembit Tenant, click Reporting in the left sidebar, then click the Workload Events tab.

  2. Filter Application Protocols to MCP, and optionally filter User to narrow the results.

  3. Open the event and find requestId under application.mcp.contentSecurity.crowdStrikeAidr.

  4. In the CrowdStrike console, search your AIDR findings for that request ID.

To work in the other direction, start from the request ID and timestamp on a CrowdStrike AIDR finding. Filter Workload Events to the matching timespan and the MCP protocol, then match the requestId field.

For how inspection works and how to configure it, see CrowdStrike AIDR Content Security.

PostgreSQL fields appear under application.postgres. Amazon Redshift uses the same fields, under application.redshift. Request and response events carry the same fields; the value of message.type differs by direction.

Request fields:

FieldPresenceDescription
protocolVersionOptionalThe PostgreSQL wire protocol version.
userOptionalThe database user.
databaseOptionalThe target database name.
message.typeAlwaysThe protocol message type, such as Query.
tls.versionOptionalThe negotiated TLS version, such as TLSv1.2.

Response fields:

FieldPresenceDescription
protocolVersionOptionalThe PostgreSQL wire protocol version.
userOptionalThe database user.
databaseOptionalThe target database name.
message.typeAlwaysThe protocol message type.
tls.versionOptionalThe negotiated TLS version, such as TLSv1.2.
Example application.postgres block
"application": {
"postgres": {
"protocolVersion": "3.0",
"user": "analytics",
"database": "warehouse",
"message": {
"type": "Query"
},
"tls": {
"version": "TLSv1.2"
}
}
}

MySQL fields appear under application.mysql. MariaDB shares the MySQL protocol handler and uses the same event types and fields. Request and response events carry the same fields; the value of message.type differs by direction.

Request fields:

FieldPresenceDescription
protocol_versionAlwaysThe MySQL wire protocol version.
server_versionAlwaysThe MySQL server version string, such as 8.4.7.
server_thread_idAlwaysThe server-assigned connection thread ID.
userAlwaysThe database user.
message.typeAlwaysThe protocol message type, such as ComQuery.
connection_attributesOptionalClient-supplied connection metadata, such as client name and version, operating system, and program name.

Response fields:

FieldPresenceDescription
protocol_versionAlwaysThe MySQL wire protocol version.
server_versionAlwaysThe MySQL server version string, such as 8.4.7.
server_thread_idAlwaysThe server-assigned connection thread ID.
userAlwaysThe database user.
message.typeAlwaysThe protocol message type.
connection_attributesOptionalClient-supplied connection metadata, such as client name and version, operating system, and program name.
Example application.mysql block
"application": {
"mysql": {
"protocol_version": "10",
"server_version": "8.4.7",
"server_thread_id": 4182,
"connection_attributes": {
"_client_name": "libmysql",
"_client_version": "8.4.7",
"_os": "Linux",
"program_name": "mysql"
},
"user": "reporting",
"message": {
"type": "ComQuery"
}
}
}

Oracle Database fields appear under application.oracleDatabase, and Redis fields under application.redis. Both carry a single field, whose value differs between request and response events.

Request fields:

FieldPresenceDescription
message.typeAlwaysThe protocol message type. For Redis, the command name, such as AUTH. For Oracle Database, the packet type, such as Connect (NSPTCN).

Response fields:

FieldPresenceDescription
message.typeAlwaysThe protocol message type. For Redis, the status, such as OK or ERR. For Oracle Database, the packet type, such as Accept (NSPTAC).
Example application.oracleDatabase and application.redis blocks
"application": {
"oracleDatabase": {
"message": {
"type": "Connect (NSPTCN)"
}
}
}
"application": {
"redis": {
"message": {
"type": "AUTH"
}
}
}

TCP Passthrough captures traffic that doesn’t match a supported application-protocol handler. Instead of request and response events, Agent Proxy generates the connection-level event types tcp.open and tcp.close, which don’t include an application block. When Agent Proxy can identify the underlying protocol, it records it in meta.identifiedProtocol.