Porter's MCP tools have detailed descriptions in Chinese but critical structural gaps. All 7 tools have descriptions (good), but parameter schemas are inferred from the Python validation script rather than visible in actual Go source code. Input schemas show types and descriptions for most parameters, but output schemas are completely undocumented. Tool naming follows verb_noun pattern (download_start, download_wait, list_tasks) which is appropriate. However, descriptions are overly technical and assume domain knowledge of download protocols, mirrors, and digest formats, not optimized for LLM selection. No tool annotations (readOnlyHint, destructiveHint) despite clear risk classifications. Error handling guidance is absent from descriptions.

• Document output schemas for all 7 tools. Example: download_start should return {task_id: string, status: string, created_at: ISO8601}. download_status should return {task_id, status, bytes_downloaded, total_bytes, speed_bps, error_message?, checksum?}. This enables tool chaining and LLM planning.
• Rewrite descriptions in plain English, 50 - 150 characters, optimized for LLM selection. Current: '异步启动下载任务，立即返回 task_id。支持镜像候选、慢速保护...' (80+ chars of jargon). Better: 'Start an async download. Returns task_id immediately. Supports multiple mirrors, retry on slow/stalled sources, and digest verification.'
• Add tool annotations: mark download_start and zip_extract with destructiveHint=true (they write files), download_cancel with idempotentHint=true (safe to retry), and download_status/download_probe/list_tasks with readOnlyHint=true. Use MCP's built-in annotation fields.
• Add error recovery guidance to descriptions. E.g., download_start: 'Returns error if URL is invalid or unreachable. Retryable errors (timeout, slow source) are handled internally with exponential backoff. Non-retryable errors (invalid digest, permission denied) fail immediately, check allow_hosts or expect_digest.'
• Document list_tasks output schema and add pagination. Specify: returns array of {task_id, status, url, output_dir, created_at, completed_at?, bytes_downloaded, total_bytes}. If history is large, add limit and offset parameters.