cyberneticlibrary
← library

Run isolated compute jobs remotely

modalskillsetup L2★465
blockrunai/blockrun-mcp ↗
What it does

Run disposable code in isolated remote containers with GPU

Best for

Running heavy compute (ML training, data processing), GPU-dependent code, or untrusted code in safe isolation without local GPU procurement or Docker management.

Inputs
  • · [object Object]
  • · [object Object]
  • · [object Object]
  • · [object Object]
Outputs
  • · [object Object]
  • · [object Object]
Requires
  • · blockrun_modal MCP tool
  • · Modal remote container service
  • · USDC payment per execution ($0.001–0.01)
Preconditions
  • · blockrun_modal tool available in MCP
  • · USDC wallet funded
  • · Network access to Modal API
  • · Code must be deterministic (no interactive input)
Failure modes
  • · Sandbox timeout (idle eviction after 5 min default)
  • · Out-of-memory for large dataset
  • · GPU out of stock (rare; router retries)
  • · Docker image pull fails
  • · Command fails in container
Trust signals
  • · Path-based MCP tool (blockrun_modal)
  • · Explicit hardware catalog (T4→H100, pricing per config)
  • · Idle timeout prevents runaway costs (300s default)
  • · Setup command separation allows caching across execs
  • · Documented endpoint catalog with timeout/cpu/memory/gpu fields