core

MCP Tools

A collection of my custom MCP tools for day-to-day workflows.

Included tools

Tool	What it does	Status	Docs
gcloud-mcp	Start/stop/status workflows for Google Compute Engine instances, including GPU-focused use cases.	Active	Tool README
slurm-watch-mcp	Watch remote SLURM jobs over SSH and wait for queue exit or terminal completion.	Active	Tool README

Repository layout

mcp-tools/
  scripts/
    run-<tool-name>.sh
    ...
  servers/
    <tool-name>/
      src/<tool_package>/server.py
      README.md
    gce-ops-mcp/
      src/gce_ops_mcp/server.py
      README.md
    slurm-watch-mcp/
      src/slurm_watch_mcp/server.py
      README.md

Setup and usage

Detailed install, Codex/Claude config, and usage examples are documented per tool. Use the link in the Docs column in the table above.

What This Repo Tries To Do

• Keep setup simple.
• Avoid hardcoded personal project IDs, usernames, or instance names.
• Keep cluster access details environment-driven where possible.
• Support environment-driven defaults (GCP_PROJECT_ID is optional).
• Keep Local SSD stop behavior explicit.
• Stay easy to extend with more cloud MCP servers over time.

Status

• Works well for local use.
• Versioning and CI can be added as this grows.

Marketplace Alignment

This repo now includes a marketplace-style plugin/skill scaffold so the GCP workflow can be ported into fundamental-physics/marketplace with minimal restructuring:

.claude-plugin/marketplace.json
plugins/core/.claude-plugin/plugin.json
plugins/core/skills/gce-ops-mcp/SKILL.md