NodeAgent exposes 4 tools with minimal schema rigor. Tool definitions are inferred from JSX UI components rather than explicit MCP tool registration. Descriptions are present but generic (10-50 chars). Input schemas show only single parameters with basic type/description; no output schemas documented. No error handling guidance, no parameter constraints (enums, ranges), no idempotency hints. Naming follows verb_noun pattern (collect_context, search_synthesize, apply_spreadsheet_delta, write_memo) but descriptions lack WHEN/WHY context.

• Migrate tool definitions from JSX UI components to explicit MCP tool registration with formal JSON Schema. Each tool must have a name, description (50-200 chars), and complete input/output schemas.
• Expand descriptions to 50-200 characters. Include WHAT the tool does, WHEN to call it (vs similar tools), and WHAT it returns. Example: 'Gather context about the current room (participants, recent messages, topic). Call this before search_synthesize to understand what to search for.'
• Document output schemas for all tools. Specify field names, types, and whether results are paginated. Include IDs/references needed for downstream tool calls (e.g., if write_memo returns a memo_id, ensure other tools accept it).
• Add parameter constraints: enums for known values (e.g., focus: ['participants', 'messages', 'topic']), ranges for numeric params, regex patterns for strings. Replace example values with formal constraints.
• Implement error handling patterns: categorize errors as retryable/user-fixable/fatal, include the invalid value and constraint violated, and suggest recovery steps (e.g., 'Invalid focus: got "xyz", must be one of: participants, messages, topic. Try search_synthesize() instead.').