Summarize this article with:

summary

Claude Code runs locally in your terminal with no backend server needed
A Mac Mini M4 costs $599 and uses under 10W idle ($2/month electricity)
Point ANTHROPIC_BASE_URL at a gateway for multi-provider access
Pay-per-use API access costs $0.003 to $0.015 per 1K tokens
Eden AI provides 500+ models through Anthropic-compatible endpoints

Setup	Hardware	Cost	Best For
Cloud-routed agent	Mac Mini M2 (8GB)	$499	Basic coding tasks
Cloud-routed agent (recommended)	Mac Mini M4 (16GB)	$599	All-day coding
Local inference (27B models)	Mac Studio M2 Max (32GB)	$1,999	Offline-capable
Local inference (70B models)	Mac Studio M2 Ultra (64GB)	$4,999	Full local stack

Got a spare Mac collecting dust? It can run your AI coding agent around the clock. Claude Code runs in the terminal, talks to model APIs directly, and needs no server infrastructure. Pair it with a gateway for multi-provider access, and you have a self-hosted coding assistant that never depends on one vendor.

Why a Spare Mac Works

Claude Code is a terminal-based coding agent by Anthropic. It reads your codebase, makes edits, runs tests, and handles multi-file refactoring. The key point: it runs locally on your machine but calls cloud APIs for the AI model.

This means any Mac with internet access works. The Mac itself does not need a powerful GPU. The model inference happens in the cloud. Your Mac just runs the agent logic and displays results.

Hardware Requirements

Step-by-Step Setup

Step 1: Install Claude Code

Open the terminal on the Mac and install Claude Code via npm (Node Package Manager, the JavaScript package installer):

npm install -g @anthropic-ai/claude-code

‍

Step 2: Configure the Gateway

Instead of pointing Claude Code directly at Anthropic's API, point it at Eden AI. This gives you access to multiple providers and automatic fallbacks:

export ANTHROPIC_BASE_URL="https://api.edenai.run/v3/v1/messages"
export ANTHROPIC_AUTH_TOKEN="YOUR_EDEN_API_KEY"

‍

Now Claude Code sends its requests through Eden AI's gateway. The gateway routes to Anthropic by default but can switch to other providers if Anthropic has an outage.

Step 3: Set Up Automatic Start

Create a launchd plist so the agent starts when the Mac boots:

# Save as ~/Library/LaunchAgents/com.claudecode.agent.plist
# Then load with: launchctl load ~/Library/LaunchAgents/com.claudecode.agent.plist

‍

Avoiding Provider Lock-In

The gateway approach means you are never tied to one provider. If Anthropic raises prices, you change one environment variable. If they have an outage, the gateway's fallback routes to another provider automatically.

Eden AI's /v3/v1/messages endpoint is Anthropic-compatible. It accepts the same request format Claude Code sends and returns the same response format. The translation happens transparently.

Cost Comparison

For a developer doing 5 hours of AI-assisted coding daily, the gateway approach with pay-per-use typically costs $30 to $80 per month. That is comparable to a subscription but with no caps and access to every provider.

Running Open Source Models Locally

If you want to go fully local (no cloud API calls), Macs with Apple Silicon handle open source models well. Qwen 3.8 27B runs on a Mac Studio with 32GB RAM using Ollama or llama.cpp. You can point Claude Code at a local model through a proxy that translates the Anthropic API format.

Conclusion

A spare Mac Mini is an excellent platform for a self-hosted AI coding agent. Combined with a multi-provider gateway, it gives you always-on coding assistance without vendor lock-in or usage caps.

FAQs - Run Claude Code on a Spare Mac

Can Claude Code run on a local Mac?

Yes. Claude Code runs in your terminal and talks to model APIs directly. It does not need a backend server or remote code index. You can point it at any Anthropic-compatible endpoint, including Eden AI's /v3/v1/messages endpoint.

What hardware do I need to self-host an AI coding agent?

For Claude Code with cloud API access, any Mac with 8GB RAM works because the model runs in the cloud. For local models like Qwen 3.8 27B, you need 16GB to 24GB of RAM. Apple Silicon Macs handle local inference well thanks to unified memory.

How do I avoid provider lock-in with Claude Code?

Set ANTHROPIC_BASE_URL to point at a gateway like Eden AI instead of Anthropic's direct API. This way, the same Claude Code instance can switch between providers without configuration changes. If Anthropic has an outage, the gateway routes to a fallback.

Is self-hosting Claude Code cheaper than a subscription?

It depends on usage. Claude Pro costs $20/month with usage caps. API access costs about $0.003 to $0.015 per 1,000 tokens. If you use more than 2 million tokens per month, self-hosting with pay-per-use API access costs less.

Can I use open source models with Claude Code?

Not directly. Claude Code expects Anthropic's API format. But you can run a proxy that translates Anthropic requests to open source models. Eden AI's /v3/v1/messages endpoint handles this translation, letting Claude Code work with any provider in the catalog.

What is the best spare Mac for a coding agent server?

A Mac Mini M2 ($599) or M4 ($599) with 16GB RAM handles cloud-routed coding agents well. For local inference, a Mac Studio M2 Max with 32GB ($1,999) runs 27B models smoothly. The key is having a dedicated machine that runs 24/7 without interrupting your main work.

Last updated onAugust 25, 2026

Clément Moreau

Clément Moreau is a Developer Relations specialist at Eden AI, where he designs practical AI workflows and helps developers integrate models from more than 30 providers through a unified API. His expertise spans OCR, NLP, speech-to-text, computer vision, API design, model orchestration, performance monitoring, and cost optimization.

Run Claude Code on a Spare Mac: Self-Hosted AI Agents