Skip to content
SciCompute

MCP for agentic science

Let your research agent run heavy experiments.

SciCompute is building the MCP layer that lets agents launch, monitor, and resume long-running scientific compute on infrastructure you control, without constant supervision. Starting with RunPod.

In development

The agent execution loop

  1. Prepare

    Local tools + Docker

    The agent writes the code and manifest, locks dependencies, and validates the Pack locally in Docker.

  2. Run

    SciCompute MCP

    The agent finds compatible compute and submits a bounded Run without handling provider credentials.

  3. Continue

    Research workflow

    The agent resumes long jobs across sessions and returns verified Results and provenance to the research workflow.

One agent-facing contract for GPU and CPU workflows in Python, R, Julia, C/C++, and Fortran.