Loading ToolBench
MCP server for code indexing, symbol analysis, dependency tracking, and delta comparison across multiple programming languages.
Scoring was not performed
No output schema documentation visible for any tool. LLMs cannot reason about what fields to expect in responses, breaking tool chaining and forcing agents to guess at field names for downstream lookups.
Parameter 'direction' in dependency_graph is described as a string but should be an enum ('dependencies', 'dependents', 'both'). Free-form strings invite hallucinated invalid values.
Numeric parameters lack JSON Schema min/max constraints. maxDepth (1-20) and depth (1-50) are documented in descriptions but not enforced in schema, leaving validation to runtime and risking invalid agent calls.
Destructive tool 'invalidate_cache' has no confirmation step or dry-run pattern. Agents may accidentally invalidate production caches. Critical cache-busting operations should require explicit confirmation.
Error responses in visible code (ChangesSince method) use inconsistent formats: sometimes SerializeError helper, sometimes inline JSON with 'Error' and 'Code' keys. Inconsistency confuses agents about which fields to check.
Parameter 'includePatterns' and 'excludePatterns' in index_project are described as arrays but lack documentation of expected glob pattern format or examples. LLMs may pass invalid patterns.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 0 | 2025-06-18+ | v2 |
| 2026-03-09 | F | 45 | - | v1 |