Three tools with complete input schemas and descriptions. Naming follows verb_noun pattern (analyze_, get_, compare_). Descriptions are detailed (150-250 chars) and explain WHAT/WHEN/HOW. However, output schemas are entirely undocumented, LLMs cannot plan downstream calls or extract structured data. Parameters lack type constraints (enums, ranges). Error handling is absent from tool specs. No tool annotations (readOnlyHint present in metadata but not in schema). Missing pagination guidance for get_evidence despite returning 'evidence rows'.

• Document output schema for each tool. Example for analyze_company: {type: 'object', properties: {why_now_score: {type: 'number'}, buying_intent_score: {type: 'number'}, signals: {type: 'array', items: {type: 'object', properties: {kind: {type: 'string'}, count: {type: 'integer'}}}}, summary: {type: 'string'}}}
• Convert signal_kind description to formal enum in inputSchema: {type: 'string', enum: ['hiring', 'funding', 'product', 'news', 'expansion', 'competitor', 'review', 'pricing']}
• Add pagination to get_evidence: limit (1-100, default 20), offset (default 0). Document that results are capped at 50 per request.
• Add error handling guidance: 'If company not found, returns 404 with available similar names. Retry with corrected name or call search_companies() first.'
• Add readOnlyHint: true to all three tools in TOOLS_SPEC to formally declare they do not modify state.