10 tools with consistent naming (verb_noun pattern: extract, render, screenshot, links, page_meta, french_company, french_geocode, dns_records, email_validate, free_preview_extract). All have descriptions (avg ~100 chars, within baseline 34-392). Input schemas present with type declarations and parameter descriptions. However, output schemas are NOT documented in source, critical gap for LLM planning. No error handling guidance visible. Tool descriptions lack WHEN/WHY context and dependency hints. Parameters are minimal (1-2 per tool) but lack constraints (enums, ranges, format specs). No tool annotations (readOnlyHint, idempotentHint). Descriptions are functional but generic, e.g., 'Extract the main content of a web page as clean markdown' lacks guidance on when to use extract vs free_preview_extract or render.

• Document output schemas for all tools. Example for french_company: {type: 'object', properties: {siren: {type: 'string'}, siret: {type: 'string'}, name: {type: 'string'}, officers: {type: 'array', items: {type: 'object', properties: {name, role}}}, naf_code: {type: 'string'}, hq_address: {type: 'string'}, status: {type: 'string'}}}.
• Add WHEN/WHY guidance to descriptions. E.g., extract: 'Use for clean markdown extraction of article/blog content. For raw HTML, use render(). For quick preview without JS execution, use free_preview_extract().'
• Add parameter constraints. url: {type: 'string', pattern: '^https?://', description: 'Public http(s) URL'}. q in french_company: {type: 'string', description: 'Company name or SIREN (9 digits) or SIRET (14 digits)'}.
• Add error recovery hints to descriptions. E.g., email_validate: '...Returns valid/invalid. If domain MX check fails, the domain may be misconfigured, suggest user verify domain ownership.'
• Add tool annotations to all tools. Mark all with readOnlyHint: true and idempotentHint: true to signal safety to agents.