poe_intel provides 5 tools with clear naming (verb_noun pattern: poe_parse_build, poe_price_check, poe_currency_rates, poe_meta_overview, poe_build_cost). All tools have descriptions (avg 120 chars, within 10-1024 range). Input schemas are present with type definitions. However, parameter descriptions are minimal or missing context. Output schemas are not documented, the code returns formatted strings rather than structured objects. Error handling is basic (try/except with generic messages). No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite all being read-only operations. The build analysis output is human-readable text, not machine-parseable JSON, which forces LLMs to parse unstructured content.

• Convert all tool outputs from formatted strings to structured JSON objects. Return {build_summary: {class, level, ehp, ...}, offense: {dps, crit, ...}, defense: {...}, issues: [...]} instead of markdown text. This enables LLMs to extract and chain data without parsing.
• Add detailed parameter descriptions with format constraints. E.g., 'item_name (string): Exact or partial item name as it appears on poe.ninja (e.g., "Kaom's Heart", "Tabula Rasa"). Partial matches are supported.'
• Add readOnlyHint: true to all tool definitions in the MCP server registration. This signals to agents that these tools are safe to call repeatedly without side effects.
• Implement structured error responses. Return {error: true, code: 'ITEM_NOT_FOUND', message: 'No results found for "xyz" on poe.ninja. Try a partial name or check the item type.', recovery: 'Call poe_meta_overview to see trending items.'} instead of plain strings.
• Document poe_currency_rates output schema. Specify that it returns {chaos_to_divine: number, timestamp: ISO8601, league: string} and explain when to call it (before pricing builds to convert between currencies).

Inferred effective spec: <=2025-11-25.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry