Stack Guardian
Health checks for a home AI stack: Ollama nodes, NVIDIA GPUs and tailnet peers
- Platforms
- CLI, Linux, macOS
- Runs
- Locally
- Version
- 0.4.1
- Listing updated
- 4 days ago
Similar MCP servers
About Stack Guardian
Stack Guardian is an MCP server for developer tools. It runs locally on your machine and is started by your AI client.
Instead of guessing why a model is slow on a multi-machine local setup, it collects a snapshot of your nodes (Ollama or OpenAI-compatible endpoints, loaded models, NVIDIA GPUs on the local machine, tailnet peers) and runs 14 small Python rules against it. Findings carry a severity, evidence and a confidence grade: one run is at most LIKELY and the same finding on a second run becomes VERIFIED. Typical findings include an unreachable Ollama endpoint, an offline tailnet peer or a model about to unload.
It ships as a CLI whose doctor command exits 0, 1 or 2, a read-only MCP stdio server, a small HTTP endpoint with Prometheus metrics and a DeepSeek Harness plugin. A finding may offer a typed fix that you run on purpose; fixes are dry runs unless --apply is passed. It uses only the standard library on Python 3.11+, and is not affiliated with the StackGuardian company.
Install
Run in a terminal
claude mcp add stack-guardian --env STACK_GUARDIAN_NODES='<STACK_GUARDIAN_NODES>' -- uvx stack-guardianAdd to claude_desktop_config.json
{
"mcpServers": {
"stack-guardian": {
"command": "uvx",
"args": [
"stack-guardian"
],
"env": {
"STACK_GUARDIAN_NODES": "<STACK_GUARDIAN_NODES>"
}
}
}
}Add to ~/.cursor/mcp.json
{
"mcpServers": {
"stack-guardian": {
"command": "uvx",
"args": [
"stack-guardian"
],
"env": {
"STACK_GUARDIAN_NODES": "<STACK_GUARDIAN_NODES>"
}
}
}
}Add to .vscode/mcp.json
{
"servers": {
"stack-guardian": {
"type": "stdio",
"command": "uvx",
"args": [
"stack-guardian"
],
"env": {
"STACK_GUARDIAN_NODES": "<STACK_GUARDIAN_NODES>"
}
}
}
}Add to ~/.codeium/windsurf/mcp_config.json
{
"mcpServers": {
"stack-guardian": {
"command": "uvx",
"args": [
"stack-guardian"
],
"env": {
"STACK_GUARDIAN_NODES": "<STACK_GUARDIAN_NODES>"
}
}
}
}Add to your client's MCP settings (most use this shape)
{
"mcpServers": {
"stack-guardian": {
"command": "uvx",
"args": [
"stack-guardian"
],
"env": {
"STACK_GUARDIAN_NODES": "<STACK_GUARDIAN_NODES>"
}
}
}
}Using several servers? Build one config file with the MCP config generator.
Configuration
STACK_GUARDIAN_NODESPath to your nodes JSON file (see examples/nodes.json)Required
What it does
- Fourteen checks
- Rules covering endpoints, models, GPUs and peers.
- Confidence grades
- LIKELY after one run, VERIFIED when seen twice.
- Exit codes
- The doctor command exits 0, 1 or 2.
- Read-only MCP
- The MCP server never records state.
- Prometheus metrics
- An HTTP endpoint serves a snapshot and metrics.
- Dry-run fixes
- Fixes apply only with --apply and leave an audit line.
Good for
- Find why a local model is slow
- Detect an offline tailnet node
- Warn when a loaded model is about to unload
- Export stack health to Prometheus
Stack Guardian FAQ
What does it need?
Python 3.11 or newer and no runtime dependencies.
Does it support AMD GPUs?
The README says NVIDIA only, and only on the machine running it.
Can I try it without a stack?
Yes, run doctor against the example snapshot, which exits 2 on purpose.
Does it fix things automatically?
No, fixes run only when you request them.
How do I add Stack Guardian to Claude Code?
Run this command in your terminal: claude mcp add stack-guardian --env STACK_GUARDIAN_NODES='<STACK_GUARDIAN_NODES>' -- uvx stack-guardian
What are the best alternatives to Stack Guardian?
The closest alternatives to Stack Guardian are Guard, Stalemodel, AIOps Field Notes, LLM Preflight and Paw Pulse.
Where this comes from
- Description: written from the project's README.
- Install steps: from the official MCP Registry entry.