AgentAnycast Node provides 7 tools with basic schemas and descriptions. All tools have non-empty descriptions (range 80-180 chars, within baseline 34-392). All input schemas are present with type definitions. However, parameter descriptions are sparse or missing entirely. Most parameters lack detail on format, constraints, or valid values. No output schemas are documented. Error handling guidance is absent. Tool naming follows verb_noun convention (get_, list_, send_, discover_) which is good, but composition could be tighter, send_task and send_task_by_skill overlap significantly. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) are present despite clear risk classifications in the metadata.

• Document output schemas for all 7 tools. Specify fields returned (e.g., get_node_info returns {peer_id: string, did: string, listen_addresses: string[], connected_peer_count: number}). This enables LLMs to chain tools and extract required IDs.
• Expand parameter descriptions with format and constraint details. E.g., 'target' in send_task: 'PeerID (base58-encoded), skill name (lowercase alphanumeric), or HTTP URL (https://...)'. 'skill' in discover_agents: 'Lowercase alphanumeric skill identifier (e.g., translate, summarize, code-review).'
• Consolidate send_task and send_task_by_skill into a single send_task tool with optional by_skill boolean parameter. Reduces LLM confusion and simplifies the interface.
• Add error handling guidance to tool descriptions. E.g., discover_agents: 'Returns empty list if no agents offer the skill. Consider broadening the skill query or checking network connectivity.' send_task: 'Returns error if target peer is offline or unreachable. Retry after a delay or use discover_agents to find an alternative.'
• Add tool annotations in MCP registration. Mark send_task and send_task_by_skill with destructiveHint=true (tasks are sent and cannot be unsent). Mark all get_* and list_* tools with readOnlyHint=true. Mark get_task_status with idempotentHint=true.