Loading ToolBench
Auto-generate AI-ready context from your codebase. One command → CONTEXT.md ready to paste into Claude, Cursor, ChatGPT or any MCP agent. Full MCP server mode included.
Six tools with clear verb-noun naming (get_*, search_*) and basic descriptions. However, critical gaps undermine quality: most tools lack parameter descriptions entirely, output schemas are undocumented, and error handling is absent. The search_components tool has a well-formed schema with enum constraints, but the other five tools have empty required arrays and no parameter documentation. No evidence of pagination, result limits, or recovery guidance. Descriptions are adequate (50-150 chars) but lack WHEN/WHY context. This is a typical community MCP server with functional but incomplete definitions.
Return the full AI-ready CONTEXT.md for the project. Use this as your primary source of truth about the codebase.
Return naming conventions, test co-location policy, path aliases, and barrel export patterns.
Return recent git commits, current branch, and uncommitted-change status.
Return detailed stack information: framework, language, styling, state management, testing tools, build setup.
Return the folder tree of the project (3 levels deep, ignoring node_modules and build artifacts).
Search the component/module inventory by name or type.
Five of six tools have empty input schemas (no properties, empty required array). Tools get_context, get_stack, get_structure, and get_conventions accept no parameters but provide no documentation of what they return. This violates the documented output schema requirement.
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract fields without knowing the response structure. This forces agents to guess at field names and types.
get_git_history accepts a 'limit' parameter (number, default 15, max 50) but the description lacks validation rules. No guidance on what happens if limit exceeds 50, or if negative values are passed. Missing actionable error handling.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 53 | 2026-07-28+ | v2 |
search_components has well-formed enum for 'type' parameter, but 'query' parameter lacks constraints. No guidance on case sensitivity, minimum length, or what happens if query matches nothing. Description says 'case-insensitive' but no error recovery guidance.
No pagination, result limits, or batching. Tools like get_context and get_structure could return very large responses (full codebase context, deep folder trees). No documented limits or pagination parameters to prevent context window exhaustion.