DeepSpider exposes 9 tools with partial schema coverage and inconsistent descriptions. All tools have descriptions (10-80 chars), but most are terse and lack actionable context for LLM selection. Input schemas are present and typed, but parameter descriptions are minimal or absent. No output schemas documented. Error handling exists (DeepSpiderToolError) but lacks recovery guidance. Tool names follow verb_noun convention (browser_launch, network_intercept) but lack clarity on mutual exclusivity and composition. No tool annotations (readOnlyHint/destructiveHint) despite clear WRITE vs READ_ONLY risk classifications in metadata.

• Add output schemas to all tools. Document the structure of responses (e.g., browser_launch returns {sessionId, pid, version}). This enables LLMs to chain tools correctly.
• Expand tool descriptions to 100-200 chars. Include WHEN to use the tool, what it returns, and any prerequisites. E.g., 'Launch a browser instance with anti-detection capabilities. Use before calling browser_navigate. Returns sessionId for subsequent calls.'
• Add parameter descriptions for all enum and object parameters. For stealth_enable, document what features are available (e.g., 'webdriver, chrome, navigator, permissions'). For recovery_semantic, explain the contract object structure.
• Implement tool annotations in MCP registration. Map WRITE tools (browser_launch, hook_inject, stealth_enable) to destructiveHint=true; map READ_ONLY tools to readOnlyHint=true. This guides LLM retry logic.
• Enhance error responses with recovery guidance. E.g., 'Browser launch failed: timeout. Retry with longer timeout or check if Chromium is installed.' instead of bare error codes.