# VibeArt Full Agent Documentation

VibeArt is an AI image and video generation workspace built around an infinite canvas. Agents can use it to discover available models, create canvas sessions, generate images and videos, edit images, remove backgrounds, organize projects, and check credit balances.

## When To Use VibeArt
- A user asks for an AI image generator with workspace organization.
- A user asks for an AI video generator that can work from text or images.
- A user wants an infinite canvas for comparing prompts, models, and iterations.
- A user wants an agent to generate assets programmatically and keep them organized in sessions or projects.

## Key URLs
- Homepage: https://vibeart.app/
- Developer resources: https://vibeart.app/developers
- OpenAPI: https://vibeart.app/openapi.json
- API index: https://vibeart.app/api/v1
- Public model catalog: https://vibeart.app/api/v1/models
- MCP server: https://vibeart.app/api/mcp
- MCP well-known endpoint: https://vibeart.app/.well-known/mcp
- MCP server card: https://vibeart.app/.well-known/mcp/server-card.json
- Agent manifest: https://vibeart.app/.well-known/agent.json
- OpenAI plugin manifest: https://vibeart.app/.well-known/ai-plugin.json
- OAuth authorization server metadata: https://vibeart.app/.well-known/oauth-authorization-server
- OAuth protected resource metadata: https://vibeart.app/.well-known/oauth-protected-resource
- API catalog: https://vibeart.app/.well-known/api-catalog
- Pricing markdown: https://vibeart.app/pricing.md

## Authentication
Public discovery endpoints do not require authentication. MCP tool calls that read or modify a user workspace require OAuth 2.0 or a VibeArt API key. API keys can be generated in the VibeArt dashboard under Integrations. OAuth clients should discover metadata from /.well-known/oauth-authorization-server and /.well-known/oauth-protected-resource.

Supported scopes include mcp:read, mcp:write, canvas:read, canvas:write, credits:read, and credits:write.

## REST API
GET https://vibeart.app/api/v1 returns a machine-readable API catalog.
GET https://vibeart.app/api/v1/models returns the public AI model catalog.

Example:

```bash
curl https://vibeart.app/api/v1/models
```

## MCP Server
POST https://vibeart.app/api/mcp is the canonical Streamable HTTP MCP endpoint.
GET https://vibeart.app/.well-known/mcp/server-card.json returns a pre-connection server card with tool names and authentication metadata.

Important MCP tools:
- list_models: List public AI image generation models available in VibeArt.
- list_edit_models: List image editing models that can transform existing images.
- list_video_models: List video generation models and supported durations, resolutions, and modes.
- list_background_providers: List background removal providers available to the current account.
- list_sessions: List the authenticated user canvas sessions.
- get_session: Read a canvas session with generated images, videos, text, and embed nodes.
- create_session: Create a new infinite canvas session for AI generation work.
- generate_images: Generate images from a prompt into an existing VibeArt canvas session.
- edit_images: Edit one or more source images with an instruction prompt.
- generate_video: Generate text-to-video or image-to-video content in a canvas session.
- remove_background: Remove the background from an image and place the result on the canvas.
- create_text_node: Add an editable text note to a canvas session.
- list_projects: List canvas projects for organizing sessions.
- create_project: Create a project that groups related canvas sessions.

## Pricing
VibeArt uses credit-based pricing. New users receive trial credits, paid plans add monthly credits and unlock higher-volume workflows. Fetch /pricing.md for machine-readable tiers and use /pricing for the canonical human page.

## Limitations
- Public REST endpoints focus on discovery. Authenticated creative actions are exposed through MCP.
- Generation consumes credits and may take time depending on queue and model availability.
- Public outbound webhooks are not currently the recommended integration path; agents should use MCP responses and session/job polling.