Skip to content

lookup-engg/velo

v0.1.0LicenseRef-Proprietary

Create product demos, feature walkthroughs, customer onboarding videos, and internal training content. Velo turns your ChatGPT conversations, documents, and company knowledge into on-brand videos.

GitHubOpen Website

Velo for ChatGPT and Codex

OpenAI's plugin package for Velo. One package serves ChatGPT and Codex. OpenAI never reads this repo: we build a ZIP from this folder and upload it at platform.openai.com/plugins. This README stays out of the ZIP.

FileWhat it is
plugin.jsonManifest in the Agent Plugins format. extensions.com.openai holds the listing, the review test cases and the release notes.
mcp.jsonThe Velo MCP server, https://api.usevelo.ai/mcp. No auth block: ChatGPT reads our OAuth metadata and registers itself (dynamic client registration). OpenAI documents an extensions.com.openai.auth override, but the Agent Plugins schema rejects it, and we don't need it.
skills/velo-video/The Claude plugin's skill and its prompt library (references/prompt-library.md), with provider-neutral wording, and uploads that pass files attached in the chat straight to upload_file. Keep both in step with plugins/claude: only the upload step differs.
assets/velo-icon.pngThe Velo icon, a 480 x 480 PNG, used as both logo and composerIcon. OpenAI needs square icons of at least 48 x 48. The Claude plugin has its own copy of the same file.

Try it before it's listed

  • ChatGPT: in ChatGPT Plugins, choose +, then Add custom MCP server. Enter https://api.usevelo.ai/mcp, choose OAuth, then Create as a plugin. Tools that change things, such as create_video, need a Business, Enterprise or Edu workspace with developer mode on (set by a workspace admin). After a backend deploy that changes tools, press Refresh on the connection.
  • Codex: add the repo with a sparse checkout of only the Codex marketplace and this folder, then install and test from the ChatGPT desktop app: 
    codex plugin marketplace add lookup-engg/velo-mcp --sparse .agents/plugins --sparse plugins/openai
    Without --sparse, Codex also reads the Claude marketplace (.claude-plugin/marketplace.json, which OpenAI supports as a legacy format), and would list the Claude plugin beside this one under the same name, velo.

Build and submit

scripts/build-openai-zip.sh   # prints dist/velo-openai-<version>.zip

Upload the ZIP at platform.openai.com/plugins (Upload new or existing plugin), then:

  1. MCPs → Connect: connect https://api.usevelo.ai/mcp and sign in with the reviewer account so OpenAI can scan the tools.
  2. Domain verification: the portal shows a token. Set it as the backend's domain verification env var, so https://api.usevelo.ai/.well-known/openai-apps-challenge returns it.
  3. Review information: enter the reviewer account's email and password. It must sign in without MFA, magic links or email codes, and have sample videos (the test cases look for "my latest Velo video").
  4. Submit for review, then Publish plugin once approved.

For a new version: raise version in plugin.json, build and upload again. Tool changes on the server don't need a new ZIP: OpenAI rescans them daily and reviews each new tool.

Rules to keep

From OpenAI's plugin guidelines:

  • No pricing, plans, upgrades or links to checkout in the listing, skill or tool results. Saying that a step needs more credits is allowed.
  • Every tool keeps explicit readOnlyHint, destructiveHint and openWorldHint annotations (the backend sets all three).
  • The display name stays "Velo": no "MCP" or "Plugin" next to the brand.