Strong tool naming (verb_noun pattern throughout) and comprehensive parameter schemas with proper types and descriptions. All 14 tools have clear, actionable descriptions (avg 180 chars, within 10-1024 baseline). Input schemas are well-formed with required fields and additionalProperties=false. However, output schemas are completely undocumented, LLMs cannot plan downstream calls or extract fields. Error handling is absent; no recovery guidance or categorization. Notable tools like notable_pack and get_context_pack lack examples of what fields they return. Security: API keys are properly gated (not exposed as parameters), but no audit trail or permission scope declarations visible.

• Document output schemas for all 14 tools. For each, specify the JSON structure, field types, and which fields are always present vs optional. Example: notable_pack returns {profile: {name, bio, timeline: [{day_number, event}]}, similar_people: [{slug, name}], peers: [...]}.
• Add error handling guidance to tool descriptions. E.g., 'If slug not found, try notable_search() first to find the correct slug.' Categorize errors: retryable (network timeout), user-fixable (invalid slug), fatal (quota exceeded).
• Expand parameter descriptions for optional fields. E.g., get_life_graph: 'include_event_links (boolean, default false): if true, returns edges linking events to people involved; set false to reduce response size.' Add min/max for numeric params (event_links_limit: 1 - 100, max_people: 1 - 500).
• Declare permission scopes for each tool. Add a 'scopes' field to tool metadata: notable_search requires none (public); get_life_context requires 'read:identity'; get_life_graph requires 'read:graph'; etc. This enables least-privilege agent setup.
• Move pricing and auth requirements out of descriptions into separate metadata fields. Descriptions should focus on WHAT the tool does and WHEN to use it, not cost or prerequisites. Use tool.cost and tool.requiresAuth fields instead.