AutumnMCP has basic tool definitions with schemas and descriptions, but falls short of production-grade quality. All four tools are explicitly registered with Zod schemas and descriptions, which is a good foundation. However, descriptions are minimal (10-50 characters), parameter descriptions lack depth and constraints, output schemas are not documented, and error handling is generic. The server follows verb_noun naming correctly (get_*, create_*) but lacks the contextual richness needed for reliable LLM tool selection. Notably, the create_customer tool has no output schema documentation, and none of the tools specify what they return beyond generic JSON responses. Error handling returns raw API errors without recovery guidance.

• Document output schemas for all tools. In tools/index.ts, add a 'returns' or 'outputSchema' field to each tool object. E.g., get_autumn_user should declare: returns {id: string, name: string, email: string, created_at: number, ...} with descriptions for each field.
• Expand tool descriptions to 100-200 characters following the pattern WHAT + WHEN + RETURN. Example: 'Get details for a customer by their internal ID. Use before checking billing status or entitlements. Returns customer profile with name, email, and Stripe fingerprint.' This helps LLMs decide when to invoke the tool.
• Add format constraints to parameter descriptions. E.g., 'id: The internal customer ID (required, string, 1-128 chars)' and 'email: A valid email address (required, format: RFC 5322)'. Constraints prevent LLMs from hallucinating invalid values.
• Implement actionable error messages. Modify returnError() to return: {isError: true, content: [{type: 'text', text: 'Customer not found. Verify the ID exists or call create_customer() to register a new customer.'}]} instead of raw API errors.
• Move API key to environment variable. Change process.argv parsing to: const AUTUMN_API_KEY = process.env.AUTUMN_API_KEY || process.argv.find(...)? and document in README that AUTUMN_API_KEY=secret bun run index.ts. This prevents key leakage in logs.
• Add enum constraints for feature_id in get_entitlement. If only a fixed set of features exist (e.g., 'pro-features', 'enterprise-suite'), declare feature_id as z.enum([...]).describe(...) instead of z.string(). This prevents hallucinated feature names.