Loading ToolBench
Provider-neutral context compiler and middleware for LLM chat and agent workflows, exposing context optimization tools via HTTP MCP-style interface
ContextFusion MCP server has 5 tools with basic schemas and descriptions, but significant quality gaps. All tools have descriptions (10-60 chars, below the 50-200 char baseline for LLM optimization) and input schemas with type definitions. However, descriptions are generic and lack actionable context for LLM tool selection. No output schemas documented. No error handling guidance. Parameter descriptions are minimal. Tools lack verb-noun naming clarity (e.g., 'context.search' vs 'search_memory'). No tool annotations (readOnlyHint, destructiveHint). Composition is reasonable (single responsibility per tool), but descriptions do not explain WHEN to use each tool or dependencies between them.
Run leave-one-out ablation on optimized blocks.
Compile optimized context into chat payload.
Read memory entries.
Build execution plan for a task.
Search cached memory entries.
Descriptions are too short (10-60 chars) and lack LLM-optimized context. Baseline is 50-200 chars. Descriptions do not explain WHEN to use each tool, dependencies, or expected outcomes.
No output schemas documented. LLMs cannot plan downstream calls or extract required fields without knowing response structure.
Tool names use 'context.' prefix but lack clear action verbs. 'context.search' is acceptable, but 'context.ablate' is domain-specific jargon. Names should be self-documenting for LLM intent inference.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 42 | 2026-07-28+ | v2 |
No error handling guidance. Tools do not indicate what errors are retryable, user-fixable, or fatal. No recovery suggestions provided.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). All tools are marked READ_ONLY in metadata but this is not reflected in MCP tool definitions.