MCP server that provides tools for running OpenClaude requests synchronously and asynchronously, managing jobs, handling permissions, and configuring providers and models.
16 tools with complete schemas and basic descriptions, but significant quality gaps. All tools have input schemas with proper JSON Schema structure and additionalProperties: false. Descriptions are present but generic (avg ~60 chars, well below the 194-char baseline). No parameter descriptions, LLMs cannot infer what 'sessionId', 'resume', 'provider', 'mode' mean without context. No output schemas documented. Error handling and recovery guidance absent. Naming is inconsistent: 'openclaude_run_sync' vs 'openclaude_run_async' uses verb_noun but lacks clarity on what 'run' means vs 'resume_session'. No tool annotations (readOnlyHint, destructiveHint, idempotentHint). Security concerns: 'openclaude_debug_env' exposes environment variables; no indication of secret injection or permission gates.
Approve the current pending action for a job.
Debug tool to check server environment variables.
Deny the current pending action for a job.
Check whether OpenClaude is installed and healthy.
Cancel a running job.
Fetch the final result of a completed job.
Inspect the current state of a job.
No parameter descriptions. All 16 tools lack descriptions for input parameters (e.g., 'sessionId', 'resume', 'provider', 'mode', 'jobId'). LLMs cannot infer meaning from names alone and will guess incorrectly.
No output schemas documented. Tools return JSON but LLMs have no schema for expected fields, types, or structure. Agents cannot plan downstream calls or extract data reliably.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 58 | 2026-07-28+ | v2 |
List configured provider/model presets.
Return the OpenClaude control model and tool-choice guidance.
Return the pending approval request for a job.
Inspect the currently selected provider defaults.
Continue a previous session via sync or async mode.
Start a long-running OpenClaude job and return a job_id.
Run a short OpenClaude request synchronously.
List known sessions from the MCP metadata store.
Change the default provider and optional default model.
Generic tool descriptions (avg 60 chars, baseline 194). Descriptions like 'Run a short OpenClaude request synchronously' lack context on WHEN to use this vs async, WHAT it returns, or prerequisites. Below 20-char floor for 3 tools.
No tool annotations. Missing readOnlyHint, destructiveHint, idempotentHint. LLMs cannot distinguish safe reads from destructive writes (e.g., approve_action vs job_status). Agents may retry destructive ops, causing duplicates.
Security risk: openclaude_debug_env exposes environment variables without permission checks. No indication of secret injection, audit logging, or scope declarations. Credentials could leak into agent traces.