MCP server for IBM QRadar SIEM — 4 intelligent tools covering all 728 QRadar API endpoints
Server has 4 well-structured tools with complete input schemas and descriptions. Tool names follow verb_noun convention (qradar_get, qradar_delete, qradar_discover, qradar_execute). Descriptions are detailed (100-250 chars) with examples. However, output schemas are not documented, LLMs cannot predict response structure for chaining. Error handling is absent from visible code. Parameters lack validation constraints (e.g., no enum for method in qradar_discover despite enum in schema). Security concern: qradar_host, qradar_token, qradar_version exposed as parameters, violating secret-injection pattern.
Delete resources from QRadar. Examples: - Delete from reference set: endpoint="/reference_data/sets/blocked_ips/1.2.3.4" - Delete saved search: endpoint="/ariel/saved_searches/123"
Find QRadar API endpoints and get EXACT schema. ALWAYS use this before qradar_execute. Returns: - Exact endpoint path (DO NOT GUESS - use exactly what's returned) - Required parameters with types - Request body schema if needed Examples: - Find user endpoints: search="user", method="POST" - Find reference data: search="reference_data/sets", method="POST"
Execute POST/PUT/PATCH requests. ONLY use endpoints returned by qradar_discover. DO NOT GUESS endpoints - if discover didn't return it, it doesn't exist. Examples: - Run AQL: method="POST", endpoint="/ariel/searches", params={"query_expression": "..."} - Add to set: method="POST", endpoint="/reference_data/sets/{name}", params={"value": "..."}
Fetch data from QRadar. Use for listing or retrieving resources. Examples: - List offenses: endpoint="/siem/offenses" - Get offense: endpoint="/siem/offenses/123" - Filter: endpoint="/siem/offenses", filter="status=OPEN" - List users: endpoint="/config/access/users" - System info: endpoint="/system/about"
Output schemas not documented. LLMs cannot predict response structure, breaking tool chaining and forcing agents to guess field names.
Credentials (qradar_host, qradar_token, qradar_version) exposed as tool parameters. Violates secret-injection pattern, credentials will appear in agent logs and traces.
No error handling guidance visible. Errors should categorize as retryable/user-fixable/fatal and suggest recovery steps (e.g., 'Endpoint not found. Call qradar_discover first.').
Parameter descriptions lack validation constraints. 'endpoint' should specify format (must start with /, no query strings). 'filter' should explain AQL syntax or link to docs.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 60 | <=2025-11-25 | v2 |
qradar_discover and qradar_execute descriptions assume LLM will follow 'ONLY use endpoints returned by qradar_discover', but no validation enforces this. Tool should validate endpoint exists before executing.