15 tools with consistent naming (verb_noun pattern) and basic descriptions. All tools have input schemas with typed parameters and descriptions. However, descriptions are generic and lack LLM-optimization guidance (e.g., 'Subscribe to webhook events' doesn't explain WHEN to use it vs poll/history). Output schemas are not documented. Error handling is absent, no recovery guidance, no actionable error messages. Missing parameter constraints (enums, ranges, formats). No mention of idempotency, side effects, or prerequisites. Composition is reasonable (single responsibility per tool), but chaining metadata (IDs needed for downstream calls) is not documented in responses.

• Document output schemas for all tools. E.g., subscribe() should return {subscription_id: string, source_id: string, delivery_url: string, created_at: ISO8601}. Include all IDs needed for downstream tool calls (poll, unsubscribe, replay).
• Expand descriptions to 80-150 chars with WHEN/WHY guidance. E.g., 'Poll for pending webhook events from a subscription. Use this for real-time monitoring or batch processing. Returns events in FIFO order. Requires subscription_id from subscribe().'
• Add error handling with actionable messages. E.g., delete-source should return: 'Source not found. Available sources: [list]. Did you mean one of these?' or 'Cannot delete source with active subscriptions. Unsubscribe first.'
• Constrain parameters: add enums for sourceType in register() (stripe, github, slack, etc.); add min/max for limit (1-100) and offset (0+); document config schema or provide examples per sourceType.
• Declare idempotency and side effects in descriptions. E.g., 'subscribe() is idempotent, calling twice with same sourceId and deliveryUrl returns the same subscription_id. Calling with different deliveryUrl creates a new subscription.'

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

• Relies on Stateful initialize / Mcp-Session-Id (removed; protocol is stateless) - make each request self-contained