Loading ToolBench
MCP (Model Context Protocol) server exposing workspace and GitHub tools via JSON-RPC 2.0 over HTTP. Provides git operations, file I/O, code search, and shell command execution in isolated workspaces with governance policy enforcement.
12 tools with complete JSON schemas and basic descriptions, but significant gaps in parameter documentation and error handling. Most tool descriptions are 20-80 chars (below the 194-char baseline for A+ tools). Parameter descriptions are sparse or missing entirely. No error recovery guidance, no confirmation patterns for destructive operations, and no tool annotations (readOnlyHint, destructiveHint). Naming is verb-first and clear (clone_repo, read_file, write_file), following pattern:tool conventions. However, descriptions lack WHEN/WHY context and actionable guidance for LLM selection. Security concerns: run_command and delete_workspace are high-risk with minimal safeguards documented.
Clone a GitHub repository into an isolated workspace. Returns workspace_id.
Delete a workspace directory and all its contents.
Stage all changes (or specific paths) and commit in a workspace.
Create and checkout a new git branch in a workspace.
Return `git diff HEAD` for a workspace.
Push the current branch to remote origin.
Return `git status --short` for a workspace.
Destructive operations (delete_workspace, run_command) lack confirmation patterns and error recovery guidance. No dry-run support or undo mechanisms documented.
Parameter descriptions are minimal or absent. 'workspace_id', 'path', 'cmd', 'branch' lack format constraints, length limits, or validation rules. LLMs cannot infer valid input ranges.
Tool descriptions are 25-55 chars (well below 194-char baseline). Missing WHEN/WHY context. E.g., 'Clone a GitHub repository' does not explain when to use clone_repo vs alternatives, or what workspace_id represents.
Inferred effective spec: 2026-07-28+.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 58 | 2026-07-28+ | v2 |
List files in a workspace directory.
Read a file from a workspace.
Run a shell command in the workspace. Returns stdout, stderr, exit_code.
Search for a string across all files in a workspace.
Write (create or overwrite) a file in a workspace.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint) present. LLMs cannot distinguish safe reads from irreversible writes without explicit hints.
run_command accepts arbitrary shell commands with no input validation, sanitization, or command whitelisting documented. High injection risk (command injection, path traversal).