Loading ToolBench
MCP Server that Exposes Project Context, Codebase Search, Safe Commands & API Spec Summary for Any Model Or MCP Client
This server has 5 tools with moderate definition quality. All tools have descriptions and detectable input schemas, but descriptions are brief (8-66 chars, below the 194-char baseline), parameter descriptions are minimal or absent, and output schemas are not documented. Tool naming follows verb_noun convention but lacks clarity about relationships. The most critical gap: parameter descriptions are either missing entirely or trivial ('Optional glob pattern for file filtering' is the longest), forcing LLMs to infer intent. No error handling guidance, no idempotency hints, no permission declarations. This is typical of a C/D-tier community MCP server.
Fetch and summarize OpenAPI/Swagger from URL
Return directory tree or key files list
Run an allowlisted command in project dir
Search files by pattern with optional glob
Validate or format a JSON string
Tool descriptions are under 100 characters; baseline is 194 chars. Descriptions lack WHEN-to-use context and prerequisites, forcing LLMs to guess when each tool applies.
Parameters lack descriptions. 'depth' has no guidance on what happens at each level. 'glob' is explained but 'pattern' is not (regex vs literal?). 'command' gives no allowlist examples or error guidance. LLMs cannot infer parameter semantics from names alone.
Output schemas are not documented in code. Callers cannot plan downstream tool calls or extract data types. For example, does search-codebase return file paths, line numbers, match context? Does project-structure return file names or sizes? Without documented return types, agents waste tokens reasoning about response structure.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 50 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 47 | - | v1 |
run-command lacks error handling guidance. If a command fails, the LLM gets a raw stderr or exit code with no suggestion for recovery. Should the agent retry? Call a different tool? Ask the user? No guidance means agents stall.
run-command performs writes but has no confirmation step, dry-run capability, or rollback tool. An agent could irreversibly modify the codebase (delete files, commit changes) without safeguards.
fetch-api-spec lacks timeout handling. If the remote API hangs, the tool blocks indefinitely and the agent stalls. No explicit timeout or time-limit guidance in description.
validate-json 'action' enum includes 'format' and 'minify' but does not document what each does or what the output format is (prettified JSON, minified JSON, validation report?).
No tool declares required permissions (read:filesystem, exec:command, etc.). Agents cannot verify they have authority, and audit trails are unclear about what scope each tool requires.