Server has 7 tools with explicit schemas and descriptions. Naming follows verb_noun pattern (web_status, web_observe, web_act, web_wait, web_extract, web_search, web_run). Descriptions are present (avg ~80 chars) but lack depth on WHEN to use each tool and dependencies between them. Input schemas are well-structured with types, enums, and ranges. However, output schemas are not documented, LLMs cannot predict what fields to expect from responses. Error handling is minimal; no recovery guidance provided. Tool composition is reasonable but web_run's 'steps' parameter lacks schema detail for nested objects.

• Document output schemas for all tools. E.g., web_observe should return {elements: [{id, selector, role, name, text}], total: number}. This enables agents to plan downstream calls.
• Replace web_run 'steps' array-of-objects with a discriminated union or explicit step type schema. Define what fields each step type (observe, act, wait, extract) accepts.
• Add WHEN-to-use guidance to descriptions. E.g., 'Use web_observe for a quick DOM map; use web_extract for structured data extraction with custom field definitions.'
• Add error recovery hints. E.g., web_act: 'If element not found, try web_observe to find the correct selector, or use a different action (e.g., press instead of click).'
• Define schemas for web_extract 'fields' and 'items' parameters. Provide examples of valid JSON payloads or link to a schema definition.