This server exposes 12 tools that wrap an Aionis SDK client. While schemas are present and mostly complete, descriptions are minimal (10-20 chars for most tools), parameter documentation is sparse, and error handling guidance is absent. Tool names follow a consistent aionis_* prefix but lack action verbs (begin/resume/turn are weak; should be start_session, continue_session, advance_session). Many parameters are typed as 'object' or 'unknown' with vague descriptions like 'Authority', 'Guide', 'Evidence data', LLMs cannot infer what these contain. No output schemas documented. No error recovery guidance. The server reads like an SDK wrapper rather than an agent-optimized tool interface.
Submit exact guide-attributed feedback through the SDK.
Suppress, unsuppress, rehydrate, or activate memory through the SDK.
Return the canonical SDK Agent context without an independent MCP renderer.
Check whether the configured Aionis Runtime is reachable.
Recover exact archived memory through the SDK.
Store ordinary project memory through the SDK governed observe path.
Vague parameter descriptions: 'Authority', 'Guide', 'Evidence data', 'Evidence kind' lack actionable detail. LLMs cannot infer what these objects contain or how to construct them.
Many parameters typed as 'object' or 'unknown' with no schema detail. LLMs cannot validate or construct these inputs. E.g., 'model_config', 'budget', 'context_options', 'request' lack field definitions.
No output schemas documented. Tool descriptions do not explain what fields are returned or how to use results in downstream calls. E.g., aionis_session_begin returns 'handle' but no schema for its structure.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 44 | 2025-06-18+ | v2 |
Capture a completed host action through AgentSession and return the new state handle.
Begin the canonical SDK AgentSession and return its portable handle.
Finish through SDK verifier gating; failed verification returns continuation instead of false completion.
Resume a serialized AgentSession against Runtime and return the refreshed handle.
Record the current observation and return the SDK-rendered Agent context for the same session.
Run the Runtime-owned verifier for the current AgentSession state.
Tool names lack action verbs. 'aionis_session_begin' should be 'start_session' or 'create_session'; 'aionis_session_turn' should be 'advance_session' or 'step_session'. Current names do not clearly signal intent.
No error handling guidance. Tools do not document what errors can occur, whether they are retryable, or what the LLM should do next. E.g., 'handle' parameter validation failures return generic 'must be an object' with no recovery path.