All-in-one web workspace for AI coding agents — Claude Code, OpenCode, Codex, Copilot, Qwen Code, Pi, Cline, and Cursor — with chat, terminal, git, browser preview, database client, checkpoints, and real-time collaboration
Clopen exposes 3 tools with significant definition gaps. Agent and AskUserQuestion have reasonable descriptions (80-120 chars), but the dynamic mcp__*__* tool lacks specificity, its description is generic and input schema uses additionalProperties:true with no concrete parameter guidance. None of the tools have documented output schemas. Parameter descriptions are sparse: Agent's 'description' param lacks context on why it's required; AskUserQuestion's nested question structure is underdocumented. No error handling guidance visible.
Delegate a self-contained subtask to a specialized subagent. Pass the subagent slug as `subagentType` and the task as `prompt`. Available subagents are dynamically populated from configured subagents.
Ask the user one or more multiple-choice questions when you need a decision only they can make. Blocks until the user answers in the UI.
Dynamic MCP tools from internal and external servers. Tool names follow the pattern mcp__<namespace>__<tool>. Internal servers expose tools directly as custom tools. External servers are proxied to upstream MCP servers.
Dynamic MCP tool (mcp__*__*) has no concrete input schema. Uses additionalProperties:true with no parameter documentation, making it impossible for LLMs to understand what parameters to pass or what the tool does.
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract required fields (e.g., what does Agent return? What fields does AskUserQuestion's response contain?).
Agent tool's 'description' parameter lacks explanation of its purpose. Why is a 3-5 word description required? What does the system do with it?
AskUserQuestion's nested question structure (questions array with items containing question, header, multiSelect, options) is underdocumented. No guidance on max questions, max options per question, or character limits for header/labels.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 57 | 2026-07-28+ | v2 |
No error handling guidance visible. What happens if a subagent fails? If user cancels a question? If MCP tool call times out? LLMs need recovery paths.