Loading ToolBench
Local-first MCP server for code search, code intelligence, repository comparison, and safe file edits
Four tools with complete input schemas and descriptions. batch_tool_call and warm_content_index have good descriptions (100+ chars) explaining when to use them. However, parameter descriptions are sparse: batch_tool_call's 'calls' array lacks item-level guidance; warm_content_index's 'force' parameter has minimal context. server_health and content_index_status have adequate descriptions but lack output schema documentation. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite all being read-only. Error handling is present but generic (no recovery guidance). Naming is clear and verb-forward (batch_, warm_, content_index_status, server_health).
Run a short sequence of codeloupe-mcp tools in one request when later calls depend on earlier results. Use to reduce round trips for read-only reconnaissance; maximum 20 calls and recursive batch_tool_call is rejected.
Report Tantivy content-index zone status for specific files or directories. Use before repeated text_search calls in large repositories to see whether scoped paths are ready, warming, not indexed, or outside the indexed workspace.
Check server uptime, workspace roots, and indexing health. Use before broad searches in large repos: path_index powers fuzzy/path tools, content_index powers text_search shortlisting and may only cover listed zones.
Request Tantivy content-index warming for specific scoped paths. Use before repeated literal text_search in large repositories; pass subsystem directories, then inspect statuses/warming_zones and retry when ready.
Output schemas not documented. Tools return results but LLMs cannot infer response structure (fields, types, pagination). Forces agents to guess what data is available.
No tool annotations despite all tools being read-only. Missing readOnlyHint metadata prevents agents from understanding safety properties and planning accordingly.
Parameter descriptions incomplete. batch_tool_call's 'calls' array items lack guidance on valid tool names; warm_content_index's 'force' parameter has minimal context (when to use it, what it overrides).
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 59 | <=2025-11-25 | v2 |
Error messages generic and non-actionable. 'Rate limit exceeded' and 'Tool call timed out' do not guide recovery. Missing pattern: recovery-guide.
No pagination or result limits documented. warm_content_index and content_index_status accept path arrays but do not specify max count or behavior when paths exceed limits.