Server exposes 5 tools with complete input schemas and descriptions. All tools follow verb_noun naming (discover_store, list_products, start_negotiation, send_message, read_history). Descriptions are present and actionable (avg ~60 chars), exceeding the 20-char minimum. However, output schemas are not documented in the source code, LLMs cannot infer what fields to expect from responses. Parameter descriptions are minimal (domain, product_id, next_url, message, history_url lack detail about format, constraints, or dependencies). No error handling guidance visible. No pagination support despite tools that could return large lists (list_products). Risk annotations (READ_ONLY, WRITE) are present but not mapped to tool annotations in the MCP schema.

• Document output schemas for all 5 tools. Specify field names, types, and nesting. Example: discover_store returns {protocol_version: string, supported_features: [string], discount_pct: number}.
• Expand parameter descriptions to 50-100 chars each. Include format (domain: 'FQDN of the negotiate.v1 store, e.g., shop.example.com'), constraints (product_id: 'UUID or alphanumeric identifier'), and dependencies ('Requires prior call to discover_store').
• Add pagination to list_products: include limit (default 20, max 100) and offset/cursor parameters. Return total_count and next_cursor in response.
• Map Risk annotations to MCP tool annotations: mark start_negotiation and send_message with destructiveHint=true; mark discover_store, list_products, read_history with readOnlyHint=true.
• Add error handling guidance to descriptions. Example: 'If domain is invalid, returns 404. Try discover_store(domain) first to validate. If store does not support negotiate.v1, returns 400 with error message.'