Treza

Treza plugin

Make AI videos from your coding agent. This plugin connects Cursor, Codex, Gemini CLI, Claude Code, and any client that supports Agent Plugins or MCP to Treza, so your agent can make a clip from a description, finish media you already have, and build video pipelines that render and post to YouTube and TikTok on a schedule.

• Server: https://www.trezalabs.com/api/mcp (Streamable HTTP, OAuth with dynamic client registration)
• Setup for each app: https://www.trezalabs.com/integrations
• Docs: https://docs.trezalabs.com/api/mcp-server
• Privacy: https://docs.trezalabs.com/terms-and-privacy/privacy

Install

The first connection signs you in to Treza with Google and creates your account. There is no key to paste.

Cursor. Add to Cursor, or put this in ~/.cursor/mcp.json (or .cursor/mcp.json in a project) and restart Cursor:

{
  "mcpServers": {
    "treza": { "url": "https://www.trezalabs.com/api/mcp" }
  }
}

Codex CLI.

codex mcp add treza --url https://www.trezalabs.com/api/mcp
codex mcp login treza

In the ChatGPT desktop app, Treza is in the plugin catalog ChatGPT and Codex share: open Plugins, open Treza and add it.

Gemini CLI. This repo is a Gemini CLI extension:

gemini extensions install https://github.com/treza-labs/treza-plugin

Then run /mcp auth treza inside Gemini CLI. Gemini also loads the skill in skills/.

Antigravity CLI. Add Treza to ~/.gemini/config/mcp_config.json, then type /mcp to sign in:

{
  "mcpServers": {
    "treza": { "serverUrl": "https://www.trezalabs.com/api/mcp" }
  }
}

Claude Code.

claude mcp add --transport http --scope user treza https://www.trezalabs.com/api/mcp

Then run /mcp inside Claude Code and sign in.

Cline. Follow llms-install.md, or add the URL under MCP Servers, Remote Servers, with Streamable HTTP, then click Authenticate.

Clients that only start local servers. This repo ships treza-mcp, a small Node bridge that forwards every tool, prompt, and resource to the hosted server. It needs a scoped key from the API keys page in TREZA_API_KEY:

{
  "mcpServers": {
    "treza": {
      "command": "npx",
      "args": ["-y", "treza-mcp"],
      "env": { "TREZA_API_KEY": "treza_live_..." }
    }
  }
}

TREZA_MCP_URL overrides the endpoint (staging, self-hosted). The bridge runs nothing locally; it is a proxy.

What is inside

File	Purpose
plugin.json	Agent Plugins manifest
.cursor-plugin/plugin.json	Cursor plugin manifest: the same server and skill, with the logo
mcp.json	Declares the remote Treza MCP server for plugin-aware clients
skills/treza-video-pipelines/	A skill that teaches the agent the shortest path to a video, and to price a render before it starts
gemini-extension.json	Gemini CLI extension manifest (the same remote server)
llms-install.md	Setup steps an agent such as Cline can follow on its own
bin/treza-mcp.mjs	The stdio bridge published to npm as treza-mcp. The plugin itself does not run it

Tools

Twenty-six tools your agent can call:

• Make a video: make_video turns a description into one clip in a single call. It picks the model, prepares the render, and returns the price; run_pipeline starts it once the person accepts.
• Finish media you have: list_assets browses the library, import_media adds a file (a public link, or a local file sent through create_upload_url), edit_asset runs one operation over one file (captions, upscale, trim, crop, overlay, background removal), and assemble_video joins clips over narration and a music bed, with captions.
• Pipelines: list_pipeline_templates, list_node_types, get_node_type, create_pipeline, update_pipeline, publish_pipeline, list_pipelines, get_pipeline, set_schedule_paused
• Runs: run_pipeline, get_run, list_runs, cancel_run
• Channels and publishing: list_connected_channels, connect_channel returns a link that connects a YouTube channel or TikTok account in the person's own browser, and publish_asset lays a finished video out for review before it posts
• Budget and account: estimate_run_cost, get_credit_balance, get_profile, create_api_key (OAuth connections only)

Two more, open_library and confirm_publish, belong to Treza's panels: they run when the person uses the panel, never from the model.

Read-only tools carry readOnlyHint. Renders spend prepaid credits, so make_video and estimate_run_cost quote a price first, and run_pipeline starts a quoted run once, at that price. Renders run in the background and return a run to poll with get_run.

In clients that show MCP Apps panels, such as ChatGPT and Claude, results appear as panels: the finished file, the price with a Start render button, a library to pick files from, a map of a pipeline, and the post before it goes out. Other clients get the same answers as text and links. Posting with publish_asset needs a panel; without one, publish through a pipeline that ends in a youtube-upload or tiktok-upload node.

Runs spend prepaid credits on your Treza account. Agents with a wallet can top up over x402 with no human in the loop.

MIT licensed. Support: [email protected]