Google SecOps Remote MCP ADK Agent with Gemini Enterprise OAuth Passthrough
Critical failures across all dimensions. Tool definitions are inferred from agent.py rather than explicitly registered with schemas. No input schemas visible in source code. Descriptions are present but generic and lack LLM-optimization guidance. 13 tools declared but no parameter documentation, type definitions, or output schemas found. Many tools marked 'disabled by policy' yet still exposed. No error handling patterns, recovery guidance, or confirmation mechanisms for destructive operations despite 5 DESTRUCTIVE-risk tools.
Remote Chronicle OneMCP server tools for SIEM event search, entity investigation, detection rule management, and SOAR case operations
Create a feed in Chronicle (disabled by policy)
Deactivate a parser in Chronicle (destructive, requires confirmation)
Delete a data table row in Chronicle (destructive, requires confirmation)
Delete a feed in Chronicle (destructive, requires confirmation)
Disable a feed in Chronicle (disabled by policy, destructive)
No input schemas visible in source code. All 13 tools lack documented parameter types, constraints, and descriptions. Cannot verify LLM can infer valid inputs.
5 destructive tools (delete_feed, deactivate_parser, delete_data_table_row, execute_bulk_close_case, execute_manual_action) lack confirmation/dry-run patterns. Descriptions mention 'requires confirmation' but no mechanism visible in code.
Inferred effective spec: 2025-06-18+.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 33 | 2025-06-18+ | v2 |
Enable a feed in Chronicle (disabled by policy)
Execute bulk close case operation in Chronicle (destructive, requires confirmation)
Execute a manual action in Chronicle (destructive, requires confirmation)
Get a feed from Chronicle (disabled by policy)
List feeds in Chronicle (disabled by policy)
Run a parser in Chronicle (disabled by policy)
Update a feed in Chronicle (disabled by policy)
Tool definitions inferred from agent.py rather than explicitly registered. No McpToolset registration, tool() decorators, or schema definitions visible.
Descriptions are generic and under 50 chars for most tools (e.g., 'Disable a feed in Chronicle (disabled by policy, destructive)'). Lack WHEN/WHY guidance for LLM selection.
No error handling patterns, recovery guidance, or actionable error messages visible. Destructive operations and policy-disabled tools lack clear failure modes and next-step guidance for LLM.