Server exposes 6 read-only tools with basic schemas and descriptions. All tools have input schemas and descriptions present, but quality is inconsistent. Naming follows verb_noun convention (agency_list_*, agency_get_*) which is good. However, descriptions lack depth, most are 50-100 chars, below the 194-char baseline for production tools. Parameter descriptions are minimal or absent. No output schemas documented. No error handling guidance. No tool annotations (readOnlyHint, etc.). All tools are READ_ONLY, which is a security positive, but the lack of structured error responses and recovery guidance is a significant gap.

• Add outputSchema to each tool definition documenting the response structure. E.g., agency_list_agents should declare: {type: 'object', properties: {studio: {type: 'object'}, agents: {type: 'array', items: {type: 'object', properties: {id, name, title, emoji, audit_day, specialty, tagline, profile}}}, total: {type: 'integer'}}}.
• Expand descriptions to 100-200 chars. Include WHEN to call the tool and WHAT it returns. E.g., agency_get_schedule: 'Retrieve the weekly audit schedule showing which agent is responsible for each day and their focus area. Use this to understand agent responsibilities and plan coordination. Returns an array of daily audit entries with agent_id, focus_area, and audit_day.'
• Add parameter descriptions for all params. E.g., agency_list_project_agents 'project' param: 'Project name (e.g., jenninexus, martiangames). If omitted, defaults to AGENCY_PROJECT environment variable or --project CLI argument. Leave empty to list all available projects.'
• Remove example values from descriptions (e.g., 'e.g. "src/css/theme.css"'). LLMs reuse examples literally. Instead, use format constraints: file_path should have pattern '^[a-zA-Z0-9/_.-]+$' in schema.
• Add readOnlyHint: true to all tool definitions to explicitly signal safety. This enables MCP clients to apply read-only UI affordances and prevents LLMs from treating these as potentially destructive.