Loading ToolBench
Unified MCP server for the hybrid knowledge system exposing core tools with progressive disclosure for deeper workspaces
MemroOS MCP server has significant definition quality gaps. All 8 tools have basic descriptions (50-100 chars), but input schemas are present with minimal type information. Parameters lack detailed descriptions explaining constraints, formats, or valid ranges. No output schemas are documented. Tool names use 'mcp_' prefix (non-standard verb_noun pattern). Error handling is absent, no recovery guidance, categorization, or actionable error messages. Composition is weak: tools like mcp_knowledge_save and mcp_memory_save appear to duplicate functionality without clear distinction. No security annotations visible (no permission gates, scope declarations, or audit trail patterns). This server would require substantial rework before production use.
Compile wiki documentation from knowledge sources
Recall knowledge from the hybrid knowledge system using semantic search
Save knowledge to the hybrid knowledge system with optional Artyfacts integration
Recall memories from the memory service
Save memory to the memory service
Load a skill from the skills directory
Get the tool attention catalog with capability metadata
Open a workspace to access deeper capabilities
Tool naming violates verb_noun convention. All tools prefixed with 'mcp_' (e.g., 'mcp_knowledge_save' instead of 'save_knowledge'). LLMs parse intent from the first word, 'mcp_' is a namespace, not an action verb. This reduces clarity and makes tool selection harder when many tools are available.
No output schemas documented. LLMs cannot plan downstream calls or extract required fields without knowing what each tool returns. This forces agents to guess at response structure and risks context loss between turns.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 39 | 2026-07-28+ | v2 |
Parameter descriptions are minimal or missing. 'filters' in mcp_knowledge_recall is described as 'Optional filters for search', what filters? What format? What keys are valid? LLMs cannot construct valid calls without explicit constraints.
Duplicate or overlapping tool responsibilities. mcp_knowledge_save and mcp_memory_save both persist data; mcp_knowledge_recall and mcp_memory_recall both retrieve data. No clear distinction in descriptions, agents cannot determine which to use for a given task.
No error handling or recovery guidance. Tools lack descriptions of failure modes, retryability, or what the agent should do if a call fails. A bare error response gives the LLM no path forward.