RustBrowser defines 8 tools with comprehensive input schemas and descriptions. All tools have detailed parameter documentation (avg 72 chars per param, exceeding baseline). Naming follows verb_noun convention (fetch, session_*). Schemas are well-typed with enums and constraints. However, output schemas are not explicitly documented in the source, only inferred from code. Error handling lacks recovery guidance (no 'try X instead' patterns). Session-based tools lack idempotency guarantees. No tool annotations (readOnlyHint, destructiveHint) despite clear risk profiles.

• Document output schemas for all 8 tools. For fetch/fetch_many, specify: { content: string, title: string, links?: array<{href, text}>, tables?: array<...>, actions?: array<{id, type, label}>, stats?: {raw_tokens, distilled_tokens} }. For session_* tools, specify the snapshot structure and action_tree format.
• Add tool annotations to MCP schema: fetch and fetch_many should have readOnlyHint=true; session_submit_form should have destructiveHint=true for non-GET forms; session_close should have destructiveHint=true.
• Enhance error descriptions with recovery paths. E.g., fetch: 'If timeout occurs, increase timeout_secs or reduce max_bytes. If 429 (rate limited), reduce per_host_concurrency or increase rate_limit delay.' session_navigate: 'If navigation fails, verify the URL is valid and the site is accessible. Check allow_local if targeting localhost.'
• Document idempotency for session tools. E.g., session_click: 'Repeated calls with the same session_id and action_id are idempotent, the page state after the second click is identical to after the first.' session_navigate: 'Idempotent, navigating to the same URL twice produces the same snapshot.'
• Change session_submit_form default: require confirm=true for all non-GET submissions. Update description: 'For non-GET forms, confirm must be explicitly set to true to prevent accidental destructive submissions.'