Package Exports
- @ebowwa/mcp-modal
Readme
@ebowwa/mcp-modal
Modal GPU Cloud MCP server - manage GPU workloads, training, and robot inference.
Features
- App Management: Create and manage Modal apps
- Function Management: Call and spawn Modal functions
- Robot Inference: Deploy robot inference clients on A100/H100 GPUs
- Training Clients: MjLab training on H100 GPUs
- Video2Robot: End-to-end video-to-policy training
- Volume Management: Create and manage persistent volumes
- Secret Management: Create and manage secrets
Installation
bun add @ebowwa/mcp-modalConfiguration
Set the following environment variables:
MODAL_TOKEN_ID=your-modal-token-id
MODAL_TOKEN_SECRET=your-modal-token-secretUsage with Claude Code
Add to your MCP configuration:
{
"mcpServers": {
"modal": {
"command": "bun",
"args": ["run", "@ebowwa/mcp-modal"]
}
}
}Available Tools
Health Check
modal_health- Check Modal credentials and connection status
App Management
modal_create_app- Create or get a Modal app by name
Function Management
modal_call_function- Call a deployed Modal functionmodal_spawn_function- Spawn a function asynchronously
Robot Inference
modal_deploy_robot_inference- Deploy robot inference client (A100/H100)modal_generate_action- Generate robot action from observation
Training
modal_deploy_mjlab_trainer- Deploy MjLab trainer (H100)modal_preprocess_motion- Preprocess CSV motion datamodal_train_policy- Train robot policy
Video2Robot
modal_deploy_video2robot- Deploy Video2Robot trainermodal_process_video- Process video to extract motionmodal_train_from_video- Full pipeline: video -> policy
Volume Management
modal_create_volume- Create or get a persistent volume
Secret Management
modal_get_secret- Get a Modal secretmodal_create_secret- Create a secret from key-value pairs
Image Management
modal_image_from_registry- Create image from registry tag
Client Management
modal_list_deployed_clients- List all deployed clientsmodal_cleanup_client- Clean up a deployed client
GPU Types
| Type | Use Case |
|---|---|
| T4 | Development, light workloads |
| A100 | Inference, moderate training |
| L4 | Cost-effective inference |
| H100 | Heavy training (mjlab) |
License
MIT