MCP Phase 2 is live: your agent now generates music, voice, 3D, moodboards, particles and scripts

The Phase 1 MCP endpoint let your agent publish games. Phase 2 lets your agent build them: every AI generator on the platform is now callable through the same one-endpoint, one-auth pipe.

What your agent can do now

• generate_sprite (5 to 20 credits per image; model picker)
• generate_sfx (5 credits per take)
• generate_moodboard (15 credits per image, 4 to 16 axes swept)
• generate_voice (8 credits; OpenAI TTS with steerable voice descriptions)
• generate_music (40 credits per track; async submit and poll)
• generate_3d_model (60 credits per call; image to 3D via TRELLIS; async)
• generate_particle_effect (4 credits; returns a validated Particle Designer JSON config)
• generate_godot_script and generate_unity_script (10 credits single file, 50 credits multi-file feature)
• generate_browser_game (10 quick or 30 full; returns an upload_id you can pipe straight into finish_game_upload)

Plus three utility tools, free to call

• list_pricing: full cost sheet, tiers and packs
• get_credits_balance: monthly plus pack breakdown, next refill date
• list_my_generations and get_generation: your recent AI output history

Every credit-costed tool debits the same balance your web tools use. Provider failures refund automatically. When you hit zero mid-flow, the tool returns a structured insufficient_credits error with the pack SKUs and a top-up URL, so your agent can either surface it as prose or parse the fields.