Quartermaster defines 3 tools with clear names (retrieve_tools, call_tool, list_servers) and reasonable descriptions (100-150 chars each). All tools have input schemas with typed parameters and descriptions. However, output schemas are not documented in the source code provided, parameter descriptions lack depth (e.g., no format constraints or examples of valid values), and error handling guidance is absent. The tool set is well-composed for a routing gateway, but descriptions fall short of LLM-optimization standards (should be 50-200 chars with actionable context). No tool annotations (readOnlyHint/destructiveHint) are visible.

• Document output schemas for all three tools. Example: retrieve_tools returns {tools: [{name, description, confidence_score, ...}], total_count}. call_tool returns {success, result, error_message}.
• Add parameter constraints to retrieve_tools: 'k must be 1-100 (clamped to maxK, default 8)' and 'query must be 1-500 characters'.
• Add error recovery guidance: 'If tool not found, call retrieve_tools() with a broader query. If policy denies access, check server configuration.'
• Add tool annotations: mark call_tool with destructiveHint=true (forwards writes), retrieve_tools with readOnlyHint=true, list_servers with readOnlyHint=true.
• Expand call_tool description to clarify idempotency: 'Forwards the call to the downstream server. If the call succeeds but the response is lost, retrying may cause duplicate side effects, use idempotent operations when possible.'