By koustubh25
Recommends homelab hardware (DIY, prebuilt, mini PC, SBC) based on your requirements, with live pricing and energy modeling.
npx claudepluginhub koustubh25/homelab-recommender --plugin homelab-recommenderDesigns 1–3 candidate hardware builds from a cleared requirements profile. Reasons from first principles about CPU/RAM/GPU/storage/PSU/case/cooling, justifies every choice against requirements, and produces structured parts lists for the price-scraper. Does not fetch prices.
Validates that a candidate build will actually work together. Checks socket/chipset match, RAM speed caps, PSU headroom, case clearances (GPU length, cooler height, radiator fit), PCIe lane allocation, and ecosystem constraints (e.g. Talos GPU extension support). Reads build-candidates.json and writes compatibility-report.json.
Validates a requirements profile for impossible or contradictory intersections before any build design work begins. Reads requirements.json, flags conflicts, and either clears the profile for the architect or returns it to intake for renegotiation.
Computes idle and load power draw for a build and converts to region-appropriate running costs. Reads build-candidates.json and requirements.json, writes energy-model.json with kWh/month and $/month across usage scenarios, plus energy-saving levers.
Conversational requirements-gathering agent for homelab hardware recommendations. Asks one focused question at a time to build a structured requirements profile. Use as the entry point before any build design work.
Renders the final human-readable build plan by synthesizing outputs from all upstream agents. Reads requirements, constraints, build candidates, prices, energy model, and compatibility report; produces a single markdown document the user can act on. Does not make new decisions.
Fetches live prices and buy links for a candidate build from region-appropriate retailers. Reads build-candidates.json, queries retailers, and writes priced-builds.json with min/median/max per part and source URLs. Always scrapes live — no caching.
A homelab hardware recommendation workflow that runs as a Claude Code plugin and can also be used directly from Codex. It considers DIY builds, prebuilts (Mac mini, Mac Studio, NUC, Beelink, Minisforum), used SFF workstations, and SBCs (Raspberry Pi, Rock 5, Jetson). It fetches live prices and models running costs.
It works especially well for choosing hardware for local AI inference, where tradeoffs like memory bandwidth, VRAM, power draw, noise, and upgrade path matter as much as raw CPU/GPU specs.
Live prices from major retailers (Mwave, Scorptec, Amazon, Apple, etc.) require a headless browser. Without it, the plugin falls back to lower-quality price sources. Set this up before installing the plugin.
Add to your Claude Code MCP config:
claude mcp add playwright -- npx @playwright/mcp@latest
Or add manually to ~/.claude.json:
{
"mcpServers": {
"playwright": {
"command": "npx",
"args": ["@playwright/mcp@latest"]
}
}
}
Then restart Claude Code.
/plugin marketplace add koustubh25/homelab-recommender
/plugin install homelab-recommender@homelab-recommender
/homelab-recommend
This repo can also be run directly in Codex without extra plugin packaging.
Open the repo in Codex and ask it to run the homelab recommender workflow from this repository.
Example prompt:
Run the homelab recommender.
If you want live retailer pricing in Codex, add Playwright MCP there as well:
codex mcp add playwright -- npx @playwright/mcp@latest
Then restart Codex. Without Playwright MCP, the workflow can still run, but the pricing stage will fall back to weaker sources.
.claude-plugin/.commands/, agents/, lib/, and test-fixtures/ are shared across both runtimes.homelab-run/ is the artifact directory in both runtimes.python3 scripts/validate_contracts.py
This checks the documented JSON handoff shapes across the agents, including DIY, prebuilt, and hybrid candidates.
It also validates a completed golden run fixture under test-fixtures/golden-run/, including constraint-analysis.json, compatibility-report.json, and PLAN.md.
The workflow will:
PLAN.md with parts, prices, buy links, and risksYou can change requirements at any checkpoint and the plugin will re-run only the affected stages.
The workflow is a pipeline of seven specialized agents:
PLAN.md with parts tables, buy links, energy costs, and risks. Pure synthesis — no new decisions.Defaults to Australia. Other regions are supported via lib/retailers.json and lib/electricity-rates.json — contributions welcome.
Node Hardware MCP - Comprehensive Hardware Monitoring and System Analysis for LLMs with real-time performance metrics
Uses power tools
Uses Bash, Write, or Edit tools
Share bugs, ideas, or general feedback.
Claude Code skill pack for Vast.ai (24 skills)
Use this agent when you need to bridge technical and sales requirements for B2B enterprise deals. This agent specializes in technical demos, POC development, RFP responses, solution architecture for sales, and technical objection handling. Handles complex enterprise sales cycles with technical evaluation phases. Examples:
Build vs Buy advisor that searches for existing solutions before vibe coding. Prevents reinventing the wheel by finding libraries, open source tools, and SaaS alternatives.
Upstash Context7 MCP server for up-to-date documentation lookup. Pull version-specific documentation and code examples directly from source repositories into your LLM context.
Semantic search for Claude Code conversations. Remember past discussions, decisions, and patterns.