Four tools with clear verb-noun naming (generate_image, list_models, browse_tools, free_tier_limits). All have descriptions (194 - 392 chars, within baseline). Input schemas present with Zod validation and enums. However, output schemas are undocumented, responses are text-only with no structured field definitions. Error handling is present but generic ('Unexpected response', 'Gave up after Xs'). No parameter descriptions for some inputs (e.g., 'scene' in list_models lacks guidance on when to filter). Tool composition is sound but output richness is limited.

• Document output schemas for all tools. generate_image should return {url: string, prompt: string, aspectRatio: string, watermarked: boolean}. list_models should return {models: [{id, label, anonymous, note, path}], total: number}.
• Add parameter descriptions: 'scene' in list_models should explain 'Filter by capability: text-to-image (prompt→image), image-to-image (image+prompt→image), or any (all models).'
• Implement structured error responses with recovery guidance. E.g., 'wall' error should return {error: 'quota_exceeded', reason: 'ip_daily|client', recovery: 'Sign in at https://veida.ai to lift limits'}.
• Add a query_generation_status(taskId: string) tool to retrieve results of timed-out jobs. Enables agents to check status without re-submitting.
• Enhance generate_image description with explicit timeout behavior: 'If generation exceeds timeoutSecs, returns task ID. Call query_generation_status(taskId) to check progress later.'