Pixazo API→Models→MCP
Model Context Protocol · Pixazo MCP servers live

Connect any AI agent to Pixazo

Image, video, audio, music, 3D, and upscale — available as live MCP servers for Claude, Cursor, Codex, OpenAI Atlas, and any MCP-compatible client. Sign in with your Pixazo account once; usage is billed to your Pixazo API funds wallet.

Get the Pixazo MCPOr pick a model
Recommended · The Pixazo MCP

One MCP. Every model.

A single connector that gives your agent access to image, video, audio, music, 3D, and upscale capabilities — with one curated model per category (Seedance 2.0 for video, GPT Image 2 for image, MiniMax Music, VibeVoice, Meshy 6, Topaz). A dozen high-level tools instead of hundreds of per-model ones.

https://mcp.pixazo.ai/pixazo/mcp
claude mcp add --transport http --scope user pixazo https://mcp.pixazo.ai/pixazo/mcp claude mcp login pixazo
What's inside
  • Video→Seedance 2.0
  • Image→GPT Image 2
  • Music→MiniMax Music
  • Speech→VibeVoice
  • 3D→Meshy 6
  • Upscale→Topaz (video), SeedVR (image)
Quickstart

One URL. Any agent.

The Pixazo MCP exposes 12 high-level tools across 6 capabilities through a single endpoint. On claude.ai and Claude Desktop, connect with your Pixazo account (OAuth); in CLI and IDE clients, use your Pixazo API key. Pick your client and paste the snippet — the agent handles routing.

# Claude Desktop / claude.ai — no key needed: # Settings → Connectors → Add custom connector → paste: # https://mcp.pixazo.ai/pixazo/mcp # (sign in with your Pixazo account on first connect) # Claude Code (CLI) — no key, sign in via OAuth: claude mcp add --transport http --scope user pixazo https://mcp.pixazo.ai/pixazo/mcp claude mcp login pixazo # (On a remote/SSH box: append --no-browser to get a paste-back URL) # Then ask in any chat: # > generate a sunset over a misty mountain range, then animate it slowly # (the agent picks the right Pixazo model for each step)
// Add to Cursor's mcp.json — sign in via OAuth when prompted { "mcpServers": { "pixazo": { "url": "https://mcp.pixazo.ai/pixazo/mcp" } } }
# Add to ~/.codex/config.toml — sign in via OAuth when prompted [mcp_servers.pixazo] url = "https://mcp.pixazo.ai/pixazo/mcp"
# List the 12 tools (response is SSE — the first data: line is the payload) # Raw HTTP test — uses your API key (curl can't do interactive OAuth). curl -N -X POST 'https://mcp.pixazo.ai/pixazo/mcp' \ -H 'Content-Type: application/json' \ -H 'Accept: application/json, text/event-stream' \ -H 'Ocp-Apim-Subscription-Key: YOUR_PIXAZO_API_KEY' \ -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'
What is MCP?

The plug-in standard for AI agents

Model Context Protocol is the open spec that lets any AI agent (Claude, Cursor, Codex, ChatGPT Atlas, OpenAI's SDK, custom clients) talk to external tools through a uniform, authenticated JSON-RPC interface.

One account, every model

Connect with your Pixazo account (claude.ai / Claude Desktop) or your Pixazo API key (CLI clients). Every MCP call is billed from the same Pixazo API funds wallet you use for the REST API — no duplicate billing.

Secure per-user auth

OAuth through AWS Cognito on claude.ai and Claude Desktop, or direct API-key auth in CLI clients; Azure APIM attributes every tool call to your Pixazo API account either way.

Built on the open standard

Every server speaks vanilla MCP — works in Claude, Cursor, Codex, OpenAI Atlas, and anything else that ships an MCP client. Add a server in 30 seconds.

FAQ

Pixazo MCP — common questions

Which AI models does the Pixazo MCP support?

Pixazo currently offers 14 live per-model MCP servers, drawn from a broader 60+ AI model catalog — additional per-model MCPs are being verified and enabled continuously. The pixazo umbrella covers all curated defaults through one connector.

Live per-model MCPs today:

  • Image (4): Seedream, GPT-Image, Nano Banana, Flux
  • Video (7): Sora, Veo, Runway, Kling, Seedance, Happy Horse, Gemini Omni
  • Audio (1): ACE Step
  • Upscale (2): SeedVR (image), Topaz (video)

Full catalog (per-model MCPs rolling out for each):

  • Image: Seedream, Flux, GPT-Image, Nano Banana, Recraft, Ideogram, Studio Ghibli, Z-Image, Hidream, Auraflow, SDXL, Bria, Pixelforge, Qwen Image, Reve, Longcat, Ernie, GLM, Nucleus, FireRed Image Edit, Smart Resize.
  • Video: Seedance, Happy Horse, Sora, Veo, Runway, Kling, Pika, Hailuo, Wan, P-Video, Higgsfield, GenFlare, Lucy Edit, Grok Imagine, Mochi, Vidu, Hunyuan Video, Pixverse, Veed.
  • Audio / TTS / Music: VibeVoice, ElevenLabs, MiniMax Music, Lyria, Chatterbox, XTTS, Tracks, ACE Step.
  • 3D: Meshy 6, Hunyuan, Hyper3D, Tripo3D, Trellis3D.
  • Virtual Try-On: Fashn VTON, IDM-VTON.
  • Upscale & restoration: Topaz, P-Image, Crystal Upscaler, SeedVR, GFPGAN, CodeFormer.
Browse the full per-model directory below or use the curated Pixazo MCP umbrella to access them all through one connector.

Does Pixazo MCP work with Claude, Cursor, Codex, and OpenAI Atlas?

Yes. Every Pixazo MCP server speaks vanilla Model Context Protocol, so it works with any MCP-compatible client — Claude Code, Claude Desktop, Cursor, OpenAI Codex, ChatGPT Atlas, and OpenAI's MCP SDK. Setup is one line per client — see the Quickstart tabs.

Can I generate images through MCP?

Yes — the Pixazo MCP umbrella routes pixazoGenerateImage and pixazoEditImageto GPT Image 2 by default for consistent, predictable quality. If you'd rather pick a specific model, connect directly to a per-model MCP like seedream-mcp, flux-mcp, nano-banana-mcp, ideogram-mcp, recraft-mcp, or studio-ghibli-mcp — full list in the directory below.

What about video generation? Which video models does the MCP support?

The Pixazo MCP umbrella routes all four video tools (pixazoTextToVideo, pixazoImageToVideo, pixazoReferenceToVideo, pixazoVideoToVideo) to Seedance 2.0for state-of-the-art quality with a single, predictable model choice. If you need a different model, the per-model directory has dedicated MCPs for Happy Horse, Sora, Veo, Runway, Kling, Pika, Hailuo, Wan, P-Video, Mochi, Vidu, Hunyuan Video, Pixverse, and more — each at mcp.pixazo.ai/<model>-mcp/mcp.

Does the Pixazo MCP support TTS, music, 3D, and upscaling?

Yes — full coverage across every category:
  • Speech / TTS: VibeVoice, ElevenLabs, XTTS, Chatterbox
  • Music & songs: MiniMax Music, Lyria, ACE Step, Tracks
  • 3D generation: Meshy 6, Hunyuan 3D, Tripo3D, Trellis3D, Hyper3D
  • Upscale & restoration: Topaz (video), SeedVR & Crystal Upscaler (images), P-Image, GFPGAN, CodeFormer
The Pixazo MCP umbrella exposes a high-level intent tool for each category — its pixazoUpscale routes videos to Topaz and images to SeedVR automatically; per-model MCPs give direct access if you need a specific one.

Is there an MCP for Seedance, Sora, Veo, Kling, Flux, Nano Banana, GPT-Image, ElevenLabs, or Topaz?

Yes — every model has its own dedicated MCP server. Quick reference:
  • mcp.pixazo.ai/seedance-mcp/mcp
  • mcp.pixazo.ai/sora-mcp/mcp
  • mcp.pixazo.ai/veo-mcp/mcp
  • mcp.pixazo.ai/kling-mcp/mcp
  • mcp.pixazo.ai/flux-mcp/mcp
  • mcp.pixazo.ai/runway-mcp/mcp
  • mcp.pixazo.ai/pika-mcp/mcp
  • mcp.pixazo.ai/nano-banana-mcp/mcp
  • mcp.pixazo.ai/gpt-image-mcp/mcp
  • mcp.pixazo.ai/seedream-mcp/mcp
  • mcp.pixazo.ai/happy-horse-mcp/mcp
  • mcp.pixazo.ai/elevenlabs-mcp/mcp
  • mcp.pixazo.ai/vibevoice-mcp/mcp
  • mcp.pixazo.ai/minimax-mcp/mcp (MiniMax Music)
  • mcp.pixazo.ai/topaz-mcp/mcp
  • mcp.pixazo.ai/meshy-3d-mcp/mcp
The Pixazo MCP umbrella locks each capability to one curated model — Seedance 2.0 for video, GPT Image 2 for image, MiniMax Music, VibeVoice, Meshy 6, and Topaz — so if you want a different model, use the per-model MCP directly from the list above.

How do I add the Pixazo MCP to my AI agent?

All clients authenticate via OAuth. On claude.ai / Claude Desktop, add it as a custom connector (Settings → Connectors → paste https://mcp.pixazo.ai/pixazo/mcp) and sign in. In Claude Code, Cursor, and Codex, add the server and sign in when prompted (in Claude Code: /mcp→ Authenticate). No API key needed. For raw HTTP/curl, pass your Pixazo API key as the Ocp-Apim-Subscription-Key header. Either way, usage bills to your Pixazo funds wallet. (Pixazo Platform subscription credits are separate and do not apply to MCP or REST API usage.)

A client fails at OAuth sign-in — how do I fix it?

If a client fails at sign-in, remove and re-add the server, and make sure you complete the browser sign-in. CLI/IDE OAuth is supported via auth.pixazo.ai. As a fallback, any client that supports request headers can use your Pixazo API key (api-console.pixazo.ai → API Key) as the Ocp-Apim-Subscription-Key header. Key-based usage bills to the same wallet as OAuth.

How much does the Pixazo MCP cost? Is there a free tier?

Adding any Pixazo MCP server is free — you only pay for the generations themselves, billed from your Pixazo API funds wallet at the same per-call rate as the Pixazo REST API. Many models include free trial funds. See the pricing page for current rates.

What happens if a generation fails? Am I charged?

No. If a generation fails for any reason — upstream model error, timeout, content-safety block, invalid input, and so on — the amount is not deducted from your Pixazo API funds wallet. You only pay for successful outputs. Failed calls that were briefly held are refunded automatically. You can verify every request and its final billing status in your API logs at api-console.pixazo.ai/api-logs.

What's the difference between the Pixazo MCP umbrella and the per-model MCPs?

The Pixazo MCP umbrella at mcp.pixazo.ai/pixazo/mcp is one connector with 12 high-level tools across 6 capabilities (pixazoGenerateImage, pixazoTextToVideo, pixazoTextToSpeech, and more — plus pixazoGetResultto poll async jobs) that route to a single curated model per capability — Seedance 2.0 for all four video tools, GPT Image 2 for image, MiniMax Music for music, VibeVoice for speech, Meshy 6 for 3D, Topaz (video) or SeedVR (image) for upscale. No model parameter to pick: predictable defaults that always work. The 14 per-model MCPs currently live at mcp.pixazo.ai/<slug>-mcp/mcpgive direct access to a specific model — use those when you want a different model than the umbrella default. Together that's 15 live install points today: the umbrella plus 14 per-model MCPs.