Genvid’s production system is a governed boundary your own agent drives. You use your existing tools and pipeline. Install the skills pack, point your agent at Genvid, and your agent works with your real production data: Claude Code, OpenAI Codex, Gemini CLI, Cursor, or anything else that loads skills from a directory.
Claude Code:
claude plugin marketplace add genvid-holdings/agent-skills
then:
claude plugin install genvid-skills@genvid-skills
OpenAI Codex (loads user skills from ~/.agents/skills):
git clone https://github.com/genvid-holdings/agent-skills.git
then:
mkdir -p ~/.agents/skills && ln -s "$PWD/agent-skills"/skills/genvid-* ~/.agents/skills/
Any other runtime: clone the repo and point your agent at agent-skills/skills/.
Genvid’s MCP server can be reached at mcp.genvid.com.
Claude Code:
claude mcp add --transport http genvid https://mcp.genvid.com;
then run /mcp in a session and complete the browser login.
OpenAI Codex:
codex mcp add genvid --url https://mcp.genvid.com;
then
codex mcp login genvid
if the browser login didn’t start.
The flow is standard OAuth 2.1 with PKCE and dynamic client registration, so any MCP client that supports browser login connects the same way.
The MCP server is remote and cannot read your disk, and a full-resolution file is too large to pass reliably through an MCP message payload. Binding a locally-generated file runs through the CLI, which streams it losslessly from the machine that holds it. If you only use hosted-URL providers, you don’t need it.
brew install genvid-holdings/genvid/genvid
then:
genvid login
Fork the pack, add your own skills alongside the reference ones, and fill in AGENTS.md with your studio’s conventions: naming, tightened gates, show-specific prompt rules. The reference skills keep teaching the boundary; yours layer on top.
Learn about the complete list of tools
Visit Genvid’s GitHub repository
Features
Titles