Single tool with strong naming (verb_noun: enrich_company_firmographics), comprehensive description (287 chars, within 10-1024 baseline), and complete input schema with Zod types and per-parameter descriptions. Tool annotations present (readOnlyHint, idempotentHint, destructiveHint). Output schema not explicitly documented, response is JSON.stringify of Apify actor results, structure opaque to LLM. Error handling is good (APIFY_TOKEN validation, HTTP status classification, actionable messages), but output schema documentation would elevate to 80+.

• Document the output schema explicitly. Describe the structure of the returned JSON: fields like 'employee_band', 'industry', 'hq', 'founded_year', 'revenue_estimate', 'logo', 'description', 'source_signals', 'data_completeness_score', and their types. This enables LLM to extract and chain results.
• Add explicit result limits and pagination guidance. State: 'Batch mode processes up to 10 domains per wave (batchSize param). Results are returned as a flat array. For >100 domains, split into multiple calls.' Include example of handling large batches.
• Enhance error recovery guidance. For 402 (insufficient credits), suggest: 'Upgrade your Apify plan or retry with fewer domains.' For 408 (timeout), suggest: 'Reduce batchSize or run fewer domains per call.'
• Move APIFY_TOKEN to a secure secret store (e.g., AWS Secrets Manager, HashiCorp Vault) instead of .mcp.json. Document how to inject it at runtime without exposing in config.
• Add a 'dry_run' or 'validate_domains' parameter to let LLMs test domain validity before consuming credits. This prevents wasted API calls on malformed domains.