Loading ToolBench
MCP server that exposes prompt capture and retrieval tools for Claude Code sessions. Provides search, log, touched, and session query capabilities over captured prompts stored in .prompts/ directory.
Four tools with complete input schemas and descriptions, but descriptions are minimal (10-20 chars baseline) and lack LLM-optimized guidance. All tools are read-only, which is good for safety, but parameter descriptions are sparse. No output schemas documented. No error handling guidance. Naming follows verb_noun pattern (search, log, touched, session) but lacks clarity on when to use each vs. alternatives.
Return the most recent N prompts (newest first).
Search prompts in this repo's .prompts/ store by case-insensitive substring on prompt text.
Return all records belonging to the given session_id, in chronological order.
Return prompts whose outcome.files_touched contains the given path substring.
Descriptions are too brief (10-20 chars) and lack LLM-optimized guidance. 'Return the most recent N prompts (newest first)' does not explain WHEN to use this vs. search or touched, or what the output structure contains.
No output schemas documented. LLMs cannot plan downstream operations or extract fields without knowing the response structure (e.g., what fields does a prompt record contain? Is there pagination?).
Parameter descriptions missing or minimal. 'query' in search has a description, but 'limit' in all tools lacks explanation of what happens when limit is exceeded (truncation? pagination?).
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 47 | 2026-07-28+ | v2 |
No error handling guidance. If session_id does not exist, or path substring matches nothing, what does the tool return? An empty list? An error? LLMs need recovery hints.
Tool naming lacks disambiguation. 'log' is generic (could mean write a log, retrieve logs, or stream logs). 'session' is vague (session of what? a prompt session? a user session?). Rename to 'list_recent_prompts' and 'get_session_prompts' for clarity.