Grok MCP
Google Grok video generation — Grok 2, Grok 3, Grok 3.1 (incl. fast variants and upscale).
Generate AI video with Google Grok from text or images, including high-quality Grok 3.1 and the fast variants. Supports 1080p upscale.
This extension registers the grok MCP server with VS Code so GitHub
Copilot and any other agent that speaks the Model Context Protocol
can call it directly from chat.
Quick Start
- Install this extension. VS Code registers the
grok MCP server automatically.
- Get an API key from Ace Data Cloud (Applications → API Key). New accounts include free trial credit.
- Open Copilot Chat in agent mode and ask for a video task — the extension prompts for the API key the first time and stores it in the OS keychain via VS Code's
SecretStorage.
You can rotate or remove the API key any time from the command palette:
- Grok MCP: Set Ace Data Cloud API Key
- Grok MCP: Clear Ace Data Cloud API Key
The default config talks to the hosted streamable-HTTP endpoint at
https://grok.mcp.acedata.cloud/mcp — no Python, no uvx, no local install needed.
VS Code Setup Guide
For screenshots, token setup, project-level and user-level mcp.json, and Copilot Agent Mode examples, see:
Example prompts
- "Generate a Grok 3.1 video of a vintage train pulling into a snowy station at dusk."
- "Animate https://example.com/portrait.jpg into a Grok Fast clip with subtle head turn."
12 tools available via this server.
| Tool |
Description |
grok_text_to_video |
Generate AI video from a text prompt using Grok. |
grok_image_to_video |
Generate AI video from one or more reference images using Grok. |
grok_get_1080p |
Get the 1080p high-resolution version of a generated video. |
grok_upsample |
Upsample a generated video to 1080p, 4K, or GIF. |
grok_extend_video |
Extend a Grok 3.1 video with additional content. |
grok_reshoot |
Re-render an existing video with a different camera motion. |
grok_video_objects |
Insert or remove objects in a generated video. |
grok_get_task |
Query the status and result of a video generation task. |
grok_get_tasks_batch |
Query multiple video generation tasks at once. |
grok_list_models |
List all available Grok models and their capabilities. |
grok_list_actions |
List all available Grok API actions and corresponding tools. |
grok_get_prompt_guide |
Get guidance on writing effective prompts for Grok video generation. |
Supported Models
grok-2, grok-2-fast, grok-3, grok-3-fast, grok-3.1, grok-3.1-fast
Pricing
From $0.30 per clip. Free trial credit on sign-up. See full pricing at https://docs.acedata.cloud.
Configuration
This extension implements the mcpServerDefinitionProviders contribution point
and registers a single hosted server with VS Code:
Provider id : acedatacloud.grok
Server label: Grok MCP
Server URL : https://grok.mcp.acedata.cloud/mcp
Transport : Streamable HTTP
Auth : Bearer API key from VS Code SecretStorage (or $ACEDATACLOUD_API_TOKEN)
You don't need to edit mcp.json — the extension handles registration and
token handling automatically. If you'd rather configure things by hand, the
sections below show equivalent mcp.json snippets you can use instead of
this extension.
Alternative: manual mcp.json (hosted)
{
"servers": {
"grok": {
"type": "http",
"url": "https://grok.mcp.acedata.cloud/mcp",
"headers": { "Authorization": "Bearer ${input:acedatacloud_api_token}" }
}
},
"inputs": [
{
"type": "promptString",
"id": "acedatacloud_api_token",
"description": "Ace Data Cloud API key",
"password": true
}
]
}
Alternative: local stdio (no network roundtrip)
For offline dev, air-gapped environments, or pinning to a specific PyPI
version, install uv and use:
{
"servers": {
"grok": {
"type": "stdio",
"command": "uvx",
"args": ["mcp-grok"],
"env": { "ACEDATACLOUD_API_TOKEN": "${input:acedatacloud_api_token}" }
}
}
}
uvx will download and run the latest mcp-grok on demand.
Links
License
MIT — see LICENSE.
