Task-scoped MCP least-privilege profiler with reviewed policies and Wazuh detections.
This is a test/lab fixture server with 17 tools, most of which are intentionally minimal for testing MCP protocol behavior (changing catalogs, fault injection, pagination). Tool descriptions are present but extremely terse (10-30 chars), well below the 50-200 char LLM-optimized baseline. Input schemas are properly defined with types and descriptions, but parameter descriptions are sparse. The server is designed for benchmarking and compliance testing, not production agent use, this explains the minimal descriptions. No error handling guidance, no output schemas documented, no tool annotations. Naming is clear and verb-forward (list_, read_, write_, fetch_, simulated_, delayed_, immediate_, emit_, change_, mark_, page_), which is a strength.
Changing-catalog fixture after_change.
Changing-catalog fixture before_change.
Changing-catalog fixture change_catalog.
Return text after a bounded delay for timeout and cancellation tests.
Terminate only this disposable lab child to simulate a server crash.
Emit an invalid protocol line, then terminate the disposable child.
Fetch a deterministic local advisory without using the network.
Tool descriptions are extremely terse (10-30 chars), far below the 50-200 char LLM-optimized baseline. Descriptions like 'Changing-catalog fixture change_catalog.' and 'Pagination fixture page_one_tool.' provide no actionable context for agent selection or understanding when to call the tool.
No output schemas documented. Tools return unstructured text or objects without declaring expected fields, types, or structure. LLMs cannot plan downstream calls or extract data reliably.
Inferred effective spec: 2025-06-18+.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 53 | 2025-06-18+ | v2 |
Return text immediately to prove the session remains usable.
List the safe fixture files available to the lab agent.
Create a harmless test marker in the operating-system temp tree.
Pagination fixture page_one_tool.
Pagination fixture page_two_tool.
Read a UTF-8 file from the safe lab fixture directory.
Simulate command execution without spawning any process.
Simulate credential access by returning a public canary.
Simulate outbound posting without making a network connection.
Write a bounded file only inside the disposable lab sandbox.
No error handling guidance. Tools do not document what errors can occur, whether they are retryable, or what the agent should do next. E.g., read_fixture can fail if the file does not exist, but the error message and recovery path are not documented.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). Tools like deliberate_child_exit and emit_malformed_and_exit are marked IRREVERSIBLE in the metadata but not annotated in the schema. Agents cannot infer side effects from the tool definition alone.
Parameter descriptions are minimal or absent. E.g., 'relative_path' in read_fixture is described as 'Path relative to fixture root' but does not explain constraints (no absolute paths, no escaping), format (forward slashes?), or examples. Agents must infer from the parameter name alone.