Arrakis
Rent GPUs at live prices and serve Hugging Face models, within a spend cap you set
- Pricing
- Paid
- Platforms
- API, CLI, Web
- Runs
- Remotely
- Version
- 0.1.0
- Listing updated
- 2 days ago
About Arrakis
Arrakis is a paid MCP server for cloud and infrastructure. It runs as a hosted remote server, so there is nothing to install locally.
Arrakis hands an agent a full rental loop: search the GPU market, rent a machine, run a job over SSH, and shut it down. The limits sit in the API rather than the prompt. An agent token carries read and rent scopes, a spend cap and a maximum rental length, and a capped token cannot rent without a removal time. Each rental counts at its worst case, hourly price times removal hours, and one that would pass the cap is refused.
It connects over streamable HTTP with a bearer token created in the console or with arrakis tokens create. Rentals are prepaid by the hour: the first hour is paid up front, each later hour as it begins, and the machine stops when the balance cannot pay. Reusing an idempotency key returns the same machine, so a retry never rents twice. Every tool result carries the last day's alerts, such as host maintenance notices, evictions or a balance about to run out, and no hour is charged while a host is not answering.
There are 13 tools, each labelled with its cost. Besides renting, deploy_model serves a Hugging Face model with vLLM, SGLang or llama.cpp and returns an OpenAI-compatible endpoint. A partner program lets agents earn a share of payments from customers they bring.
Install
Run in a terminal
claude mcp add --transport http arrakis https://arrakis.goodailabs.com/mcpAdd to Settings → Connectors → Add custom connector
https://arrakis.goodailabs.com/mcpAdd to ~/.cursor/mcp.json
{
"mcpServers": {
"arrakis": {
"url": "https://arrakis.goodailabs.com/mcp"
}
}
}Add to .vscode/mcp.json
{
"servers": {
"arrakis": {
"type": "http",
"url": "https://arrakis.goodailabs.com/mcp"
}
}
}Add to ~/.codeium/windsurf/mcp_config.json
{
"mcpServers": {
"arrakis": {
"serverUrl": "https://arrakis.goodailabs.com/mcp"
}
}
}Add to your client's MCP settings (most use this shape)
{
"mcpServers": {
"arrakis": {
"type": "http",
"url": "https://arrakis.goodailabs.com/mcp"
}
}
}Tools this server provides (13)
Taken from the project documentation; the running server may differ.
- search_gpus
- Lists available machines best first, with filters for GPU, count, memory, price and country.
- get_market_summary
- Shows the lowest price and supply of each GPU family.
- get_balance
- Shows the balance, running machine cost per hour, when money runs out and the token's cap.
- rent_gpu
- Rents a machine for a set number of hours and returns its SSH command.
- deploy_model
- Serves a Hugging Face model with vLLM, SGLang or llama.cpp and returns an OpenAI-compatible endpoint.
- list_images
- Lists the images a machine can boot.
- list_instances
- Lists the machines the account is running.
- get_instance
- Shows one machine's state, SSH command, removal time and host notices.
- get_events
- Lists events for the account's machines.
- terminate_instance
- Stops a machine now and ends billing.
- schedule_removal
- Moves a machine's removal time.
- get_partner_earnings
- Shows partner codes, referred customers and earnings.
- join_partner_program
- Joins the partner program for the account.
What it does
- Spend-capped tokens
- Tokens have scopes, a spend limit, a maximum rental length and an optional expiry.
- Required removal time
- Capped tokens cannot rent a machine without termination_hours.
- Model serving
- Deploys any Hugging Face model with vLLM, SGLang or llama.cpp and returns an OpenAI-compatible endpoint.
- Retry-safe rentals
- The same idempotency key always returns the same machine.
- Alerts on every result
- Tool results carry host notices, evictions and low-balance warnings.
- Event feed and webhook
- Machine events are available through get_events and a signed webhook.
- Partner program
- Agents can join and earn a share of payments from customers they refer.
Good for
- Rent a GPU for a few hours to run an evaluation suite, then terminate it
- Fine-tune a model on several GPUs with a fixed budget
- Compare current GPU prices, memory and host country before renting
- Serve an open model on a rented machine and get an OpenAI-compatible endpoint
Pricing
Paid
Hourly GPU rental at live market prices; prepaid balance
Prices change. Check arrakis.goodailabs.com before you buy.
Arrakis FAQ
Can an agent overspend?
No. A capped token refuses any rental whose worst-case cost would pass its cap, and cannot add funds.
How are rentals billed?
Prepaid by the hour at the hourly price, with the first hour paid up front and each later hour as it begins.
Is there anything to install?
No. It is served over MCP from the site's own address, with a bearer token.
What happens if a host fails?
Arrakis tells the agent at once, charges nothing while a host isn't answering and refunds hours begun after a host took a machine back.
Is Arrakis free?
No. Arrakis is a paid product.
How do I add Arrakis to Claude Code?
Run this command in your terminal: claude mcp add --transport http arrakis https://arrakis.goodailabs.com/mcp
What are the best alternatives to Arrakis?
The closest alternatives to Arrakis are Claude Code Ultimate Guide, Agent Yurt, Proofline, signafAIde and Toolkien.
Alternatives to Arrakis

Claude Code Ultimate Guide
MCP serverMCP server for searching the Claude Code Ultimate Guide from any MCP client
Runs locally- Open source
- 6.5K GitHub stars
- AI models and LLM tooling

Agent Yurt
MCP serverBack up agent memory, files and dashboards, then recover from a reviewed snapshot
Not checked yet- Remote
- Free
- Knowledge and memory

Proofline
MCP serverProduction-aware review of changes against your AWS or Google Cloud setup, via MCP
- Remote
- Freemium
- Developer tools

signafAIde
MCP serverLocal MCP server to set up, verify and check coverage of signed agent action records
Runs locally- 5 tools
- Freemium
- Security

Toolkien
MCP serverOne MCP gateway for all your MCP servers, with code mode, permissions and audit
Not checked yet- 3 tools
- Remote
- Open source
- AI models and LLM tooling

filex
MCP serverSelf-hosted file manager in one Go binary with a built-in MCP server for agents
Not checked yet- Remote
- Open source
- Files and documents