A desktop Claude Code MCP server without Claude Code CLI. Provides tools for executing code, managing jobs, and interacting with Claude Code agent runtime.
9 tools with basic schemas and minimal descriptions. All tools have input schemas with typed parameters (positive), but descriptions are extremely terse (10-50 chars, well below the 194-char baseline). No output schemas documented. Parameter descriptions are sparse or missing context. No error handling guidance. Tool names follow verb_noun pattern but lack clarity on distinctions (e.g., code_wait vs code_batch_wait vs code_continue). No idempotency hints, no permission gates, no recovery guidance. Composition is reasonable (separate tools for status/wait/result), but descriptions don't explain WHEN to use each variant or what distinguishes them.
Execute code with Claude Code agent runtime
Poll for results of multiple code execution jobs
Get results for multiple code execution jobs
Wait for multiple code execution jobs to complete
Get the chat history for a code execution session
Continue a code execution session and wait for completion
Start a continuation of a previous code execution session
Descriptions critically short (10-50 chars vs 194-char baseline). LLMs cannot determine when to select code_wait vs code_batch_wait vs code_continue without detailed context. Descriptions must explain WHAT, WHEN, and WHAT IT RETURNS.
No output schemas documented. LLMs cannot plan downstream calls or extract results. Must document what code_status, code_wait, and code_batch_result return (structure, field types, example values).
Parameter descriptions missing or trivial. 'job_id' and 'session_id' lack context (what format? how obtained?). 'timeout_seconds' has no range (1-3600?). LLMs cannot validate inputs without explicit constraints.
Inferred effective spec: 2025-06-18+.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 52 | 2025-06-18+ | v2 |
Get the status of a code execution job
Wait for a code execution job to complete
No error handling guidance. If code execution fails, times out, or job_id is invalid, what should the LLM do? No recovery paths documented. Must include 'If job not found, try code_status() first' or similar.
Tool naming ambiguity: code_wait, code_batch_wait, code_continue, code_continue_start all operate on sessions/jobs but lack clear distinction. Descriptions must explain: use code_wait for single job, code_batch_wait for multiple, code_continue to extend a session.