AI Director
CreateProjectsAI StudioLibraryAssetsShortcut PromptsSkillsMCPAPI Key
CreateProjectsAI Studio
Back to setup

Plugin & MCP guide

Connect, authorize and make your first call with Giggle in your AI assistant.

Choose how to connect

01Connect Giggle in ChatGPT

Add a remote MCP through ChatGPT Plugins and authorize with your existing Giggle account.

  1. Open ChatGPT Plugins and select Add custom MCP server.
  2. Enter Giggle AI Director and the server URL below. Choose OAuth with dynamic client registration (DCR) when available, enable PKCE S256, and set the scope to director:mcp. For a custom client, enter director-chatgpt and select none for token endpoint authentication.
  3. Create the plugin and finish authorization. Start a new chat and select the newly connected Giggle AI Director.
Server URL
https://giggle.pro/mcp

With the preregistered director-chatgpt client, use https://chatgpt.com/connector_platform_oauth_redirect as the callback. Prefer dynamic registration for other callbacks.

Server URL
https://giggle.pro/mcp
OAuth client ID (preregistration only)
director-chatgpt
Client secret
Leave blank
Token authentication
none
Default and base scopes
director:mcp
Authorization URL
https://giggle.pro/oauth/authorize
Token URL
https://giggle.pro/oauth/token
Authorization server base
https://giggle.pro
Resource
https://giggle.pro/mcp
PKCE
S256
OIDC
Off
Registration URL
https://giggle.pro/oauth/register
02Connect directly in ChatGPT desktop / Codex

The configuration below is a preregistered fallback for /mcp using director-local without a client secret. Prefer dynamic registration when your client supports it.

  1. Merge the configuration below into ~/.codex/config.toml. Update an existing giggle-director entry instead of adding a duplicate.
  2. Run the login command and authorize in your browser. Keep local port 8765 available for the callback.
  3. Select Restart in MCP settings, then use /mcp to find giggle-director. You can close the completed authorization page.
~/.codex/config.toml
[mcp_servers.giggle-director]
url = "https://giggle.pro/mcp"
scopes = ["director:mcp"]

[mcp_servers.giggle-director.oauth]
client_id = "director-local"
callback_url = "http://127.0.0.1:8765/callback"
callback_port = 8765
Authorize
codex mcp login giggle-director

Clients that support MCP OAuth dynamic client registration (DCR) can get a client ID automatically. Connect from MCP settings in Claude Code or Cursor and authorize in your browser. Use platform-specific preregistration for clients that require a fixed client ID.

Start with a free connection check

Send to your AI assistant
Call Giggle AI Director list_models and list_canvases and show the available models and canvases. Do not create canvases or run paid generation.

Choose one connection method per assistant to avoid duplicate plugin and direct MCP tools.

A complete creative workflow

  1. Call list_models to discover models and parameters, then list_canvases to choose an editable canvas.
  2. Use estimate_price and establish the requested work and cost before submitting generation.
  3. Keep the returned task_id and poll get_task until success, failure or timeout. Resume with the same task_id after a timeout.
  4. Do not submit paid generation again after a disconnect. Reuse the original request_id and arguments for retries; inspect the canvas for sync conflicts or review states.

MCP tool reference

Models & pricing
list_models · estimate_price
Results on your canvas
list_canvases · create_canvas
Image generation
generate_image · generate_image_from_image
Video generation
generate_video · generate_video_from_image · generate_omni_video · generate_motion_control_video · generate_avatar_video · generate_topview_avatar_video
Audio & voices
generate_speech · generate_music · clone_voice · clone_topview_voice
Uploads, helpers & tasks
upload_file · extract_avatar_subject · get_task

Frequently asked questions

How do plugins and MCP differ?

Both call the same MCP service. Manage plugins in ChatGPT Plugins and direct local connections in MCP settings or /mcp. This page does not offer a separate generation CLI.

What does it cost?

Authorization and connection checks are free. Generation uses current team credits and your member quota. Switching teams affects subsequent calls. Estimate the cost before submitting.

What if the connection drops or the backend restarts?

Keep the task_id and continue polling. A client timeout does not mean generation failed. Avoid resubmitting; successful results sync to the canvas.

Why are tools missing after connecting?

Complete OAuth, refresh the MCP connection or start a new chat, and select the correct plugin. Keep temporary tunnels running and update both server and client settings when the public address changes.