Loading ToolBench
High-performance semantic tool search for Claude Code. Save 97% tokens with intelligent MCP tool discovery.
Tool Search MCP has clear naming conventions (search, health, tools, reload, etc.) and most tools include descriptions. However, parameter descriptions are sparse or missing entirely. The search tool lacks documentation of output schema structure. Error handling guidance is absent. Parameter constraints (e.g., topK range, enum values for method) are declared in schema but not reinforced in descriptions. No tool annotations (readOnlyHint/destructiveHint) despite clear risk classifications. The rewrite-descriptions and apply-rewritten-descriptions tools have complex nested schemas that lack clarity on expected structure.
Apply rewritten tool descriptions and cache them for future use
Health check endpoint that returns server status, Ollama connection status, and tools loaded count
Reload tools from MCP configuration with optional filtering by tool names or server names
Generate rewrite prompts for tool descriptions to improve search quality
Search tools using semantic search with configurable top-K results
List available search methods and the default method
Search for tools using a specific search method (embedding, bm25, or regex)
List all available tools and their count
Output schemas not documented. Tools like search, health, tools, and search-methods return data but no schema is visible describing the response structure, field types, or pagination. LLMs cannot plan downstream operations without knowing what fields to expect.
Parameter descriptions missing or incomplete. 'topK' has a description but no guidance on valid range (e.g., 1-100). 'method' enum is declared in schema but description does not explain when to use embedding vs bm25 vs regex. 'excludeTools' and 'excludeServers' mention glob patterns but do not explain glob syntax or examples.
No tool annotations despite clear risk classifications. reload and apply-rewritten-descriptions are marked WRITE but lack destructiveHint annotation. search, health, tools, search-methods are READ_ONLY but lack readOnlyHint. Annotations help LLMs reason about safety and retry logic.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 48 | 2026-07-28+ | v2 |
Complex nested schemas lack clarity. rewrite-descriptions and apply-rewritten-descriptions accept deeply nested tool objects with input_schema properties. No documentation of expected structure, required fields, or how to construct these objects. LLMs will struggle to invoke these tools correctly.
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., reload might fail if MCP config is invalid or unreachable, no guidance provided.