Loading ToolBench
Unified MCP server suite for multi-LLM consultation with support for Gemini, Codex, Claude, Grok, Ollama, and Antigravity. Routes requests through runtime availability checks and provides code review, second opinions, and AI-to-AI collaboration.
Mixed quality across 9 tools. Strengths: all tools have descriptions (194 - 500+ chars, well above 20-char floor); input schemas present with type definitions; tool names use action verbs (ask-*, ping). Weaknesses: descriptions are verbose and LLM-unfriendly (many exceed 200 chars, burying key details); parameter descriptions lack specificity on constraints (e.g., minLength/maxLength stated in schema but not explained in prose); no output schemas documented; error handling guidance absent; security concerns (environment variable injection, no permission gates). ask-antigravity, ask-claude, ask-codex, ask-gemini, ask-grok, ask-ollama all follow similar patterns but lack actionable error recovery guidance.
Send a prompt to Google's Antigravity CLI (agy) for a subscription-backed second opinion, code review, or analysis. Requires agy >=0.1.0, installed and logged in once; unsupported versions fail before model invocation. Defaults to the gemini-3.1-pro model at high reasoning effort, falling back to gemini-3.5-flash on a rate limit (override via the ASK_ANTIGRAVITY_MODEL / ASK_ANTIGRAVITY_EFFORT env vars; run `agy models` for options); single-turn only (no multi-turn). Returns human-readable text plus a structured response.
Send a prompt to Anthropic Claude Code CLI for an independent second opinion, code review, or architecture critique. Defaults to claude-3.5-sonnet with claude-3.5-haiku fallback. Claude runs in safe mode with only Read, Glob, and Grep tools, so it can inspect context but cannot edit files or execute commands. Supports native sessions via sessionId and returns a structured AskResponse.
Send a prompt to OpenAI Codex CLI (defaults to o1-preview with automatic fallback on quota errors). Use for code review, second opinions, analysis, and AI-to-AI collaboration. Do not override the model parameter unless the user explicitly asks. Returns both human-readable text and a structured response (provider, model, sessionId, usage) via outputSchema. Calls are ephemeral by default; pass sessionId: "" on the first call to persist its Codex thread_id, then pass the returned sessionId on follow-up calls to continue the conversation.
Send a code edit request to OpenAI Codex CLI and get structured search/replace edit blocks back (via codex --output-schema). Codex reads the existing files (read-only) and proposes precise, applyable changes for Claude to apply. Use this when you want Codex to suggest specific code modifications to existing files rather than just analysis. Mirrors ask-gemini-edit.
Descriptions exceed 200 chars and bury key details. ask-antigravity (500+ chars), ask-codex (400+ chars) force LLMs to parse verbose prose instead of scanning intent. Baseline: A+ tools average 194 chars; p90=392. These exceed p90 and mix multiple concerns (model selection, fallback behavior, session handling) in one block.
No output schemas documented. Tools return 'structured AskResponse' or 'human-readable text plus structured response' but the actual response shape is not specified. LLMs cannot plan downstream calls or extract fields without knowing what fields exist.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 59 | 2025-06-18+ | v2 |
Send a prompt to Google Gemini CLI for code review, second opinions, and analysis. Defaults to gemini-3.1-pro-preview with automatic fallback to gemini-3.8-flash on quota errors. Supports native sessions via sessionId and returns a structured AskResponse.
Send a prompt to Grok through xAI API for code review, second opinions, and analysis. Requires XAI_API_KEY environment variable. Supports reasoning effort levels and returns a structured AskResponse.
Send a prompt to a local Ollama LLM for code review, second opinions, and analysis. No external provider data transfer. Supports native sessions via sessionId and returns a structured AskResponse.
Test connectivity with the Claude MCP server and check whether Claude Code CLI is installed
Test connectivity with the Antigravity MCP server and check whether agy is installed
Parameter descriptions lack actionable constraints. 'minLength: 1, maxLength: 100000' is in schema but not explained in prose. LLMs cannot read JSON Schema, they rely on description text. E.g., ask-antigravity prompt: 'The question... to send to Antigravity' does not explain the 100KB limit or why it matters.
No error recovery guidance. Tools fail silently or return raw CLI errors (e.g., 'agy not installed', 'rate limit'). Descriptions mention fallback models but do not tell LLMs what to do if a call fails. Missing: 'If rate-limited, wait 60s and retry' or 'If CLI not found, ask user to install'.
ping tools are generic and ambiguous. 'Test connectivity' and 'echo back a message' do not explain what 'connectivity' means (CLI installed? Network reachable? API key valid?). Naming 'ping' is also deprecated in MCP spec (removed in 2026-07-28).