Loading ToolBench
TDD-first AI development pipeline for Claude Code, Cursor, Codex, Gemini CLI, and Antigravity with token-saving hooks and MCP integration
This is not an MCP server. The codebase is a VS Code extension (Cursor plugin) that launches terminal commands, not an MCP tool server. Tool definitions exist only in package.json as VS Code command metadata, not as MCP tool schemas. No input schemas are visible, only command names and titles. Descriptions are minimal (10-15 chars). No structured output, error handling, or parameter validation. The server cannot be evaluated against MCP patterns because it does not implement the MCP protocol.
Open a Claude Code terminal in Cursor
Open a Codex terminal in Cursor
Open a Gemini CLI terminal in Cursor
Start a multi-tool task session with Claude, Gemini, and/or Codex terminals sharing a common AICREW_SESSION environment variable
Not an MCP server. This is a VS Code extension that spawns terminal processes. No MCP protocol implementation detected. Tool definitions exist only as VS Code command metadata, not MCP tool schemas.
No input schemas visible. Tool definitions in package.json contain only command names and titles. The startTask tool references 'sessionName' and 'tools' parameters in the description, but no JSON Schema is present.
Descriptions are under 20 characters. 'Open a Claude Code terminal in Cursor' (38 chars) is the longest; others are 10-15 chars. Descriptions lack WHEN to use, WHAT it returns, and prerequisites.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 17 | 2026-07-28+ | v2 |
No error handling or recovery guidance. If a terminal command fails or the tool is unavailable, there is no structured error response or guidance for the agent to retry or take alternative action.
No output schema documented. Tools spawn terminals but do not return structured results. An agent cannot determine success, failure, or what to do next based on the response.