MCP Quickstart

Picsart AI Playground exposes the full model catalog to AI agents as Model Context Protocol tools. Connect it to Claude Code, Cursor, Windsurf, ChatGPT, or any MCP-compatible agent and it can generate image, video, and audio across 176 models directly.

Connect

The integration works natively with MCP-capable agents. Pick your path:

Agent	How to connect
Claude Code · Cursor · Windsurf	Add the gen-ai-use Skill — it plugs the CLI into the agent
Codex (OpenAI)	Install the CLI, then add the plugin codex://plugins/picsart@openai-curated
ChatGPT / any MCP client	Connect Picsart as an MCP server — see picsart.com/gen-ai-mcp for the current endpoint and config

In all cases, authenticate once with gen-ai login.

Canonical connection details

The exact, always-current MCP endpoint and per-client config live on the official page: picsart.com/gen-ai-mcp. The tool catalog and example calls below describe what the agent can do once connected.

Tool catalog

Tool	Purpose	Spends credits
picsart_list_models	Browse the catalog (filter by mode/provider)	no
picsart_model_info	Capabilities + constraints for one model	no
picsart_model_params	JSON schema of a model's accepted params	no
picsart_validate_params	Pre-check a candidate payload	no
picsart_pricing	Quote the credit cost of a call	no¹
picsart_generate	Run any model end-to-end	yes
picsart_remove_bg	Remove an image background	yes
picsart_change_bg	Replace an image background	yes
picsart_enhance	Upscale / enhance an image	yes
picsart_vectorize	Convert a raster image to SVG	yes
picsart_credits	Current credit balance	no
picsart_upload	Upload a local file to Drive	yes
picsart_drive_list / picsart_drive_folders / picsart_drive_create_folder	Browse & organize Drive	yes

¹ picsart_pricing is a dry run — it returns cost without charging, but the lookup is per-user so it needs the token.

Recommended flow

The tools are designed to chain. A typical agent sequence:

1. picsart_list_models or picsart_model_info → pick a model
2. picsart_model_params → learn its inputs
3. picsart_validate_params → pre-check the payload
4. picsart_pricing → quote the cost
5. picsart_generate → actually run it

Example calls

Generate an image:

{ "name": "picsart_generate",
  "arguments": {
    "model": "flux-2-pro",
    "prompt": "a cat in a hat, studio lighting",
    "aspectRatio": "16:9",
    "count": 1
  } }

Generate a video with model-specific params via extra:

{ "name": "picsart_generate",
  "arguments": {
    "model": "seedance-2.0",
    "prompt": "a cat skiing down a mountain",
    "duration": 8,
    "aspectRatio": "16:9",
    "generateAudio": true
  } }

Quote a cost first:

{ "name": "picsart_pricing",
  "arguments": {
    "model": "veo-3.1",
    "prompt": "a drone shot over a snowy ridge",
    "params": { "duration": 8, "resolution": "1080p" }
  } }

Task-shaped shortcuts (auto-pick a fitting model):

{ "name": "picsart_remove_bg", "arguments": { "imageUrls": ["https://example.com/photo.jpg"] } }

Inputs reference

picsart_generate takes model and prompt (required), plus common optionals: aspectRatio, resolution, duration, count (1–8), quality, style, negativePrompt, imageUrls (image-to-X), videoUrl (video-to-X), generateAudio, enhancePrompt, and extra — a free-form object for model-specific params. Discover the exact extra shape for any model with picsart_model_params.

The result includes results: [{ url, metadata? }], an id (generation handle), and one resource_link per output URL (image or video/mp4). Clients fetch assets from URLs — never base64.

More

• Model Reference — every provider's models with MCP examples
• Pricing & Credits — how cost is computed