MCP server · Agent-ready

Start, lease and stop AWS and GCP VMs from Claude Code, Cursor or any MCP client

The Idlefy MCP server is a remote Model Context Protocol server at https://mcp.idlefy.com/mcp/. It lets an AI assistant lease, extend and release AWS and GCP machines, create dev boxes, start shared environments and read budgets and idle spend, signed in as one person through OAuth and limited to what that person may do.

https://mcp.idlefy.com/mcp/ Included in Pro

Start free Compare plans

Then connect your assistant on Pro · Last updated: October 20, 2026

any MCP client · example idlefy

you give me a mid-size dev box in frankfurt with my usual key
tool get_devbox_create_options
eu-central-1 · m7i.xlarge · key "laptop-2026"
ask Create box "alex-dev" on m7i.xlarge? AWS bills it to your account.
you yes
tool create_dev_box
done alex-dev is being created. Ask me in a few minutes and I will check.

What is the Idlefy MCP server?

The Idlefy MCP server connects an AI assistant to Idlefy through the Model Context Protocol, the open standard assistants use to call tools. Idlefy keeps cloud machines on AWS and Google Cloud stopped until someone leases them; the lease starts the machine and its end stops it. Through the server, a person asks Claude Code, Claude, Cursor or another MCP client to do that in plain words: list the machines they may use, lease one for two hours, extend it, release it, start a shared environment, or create a personal dev box.

The assistant signs in as that person with OAuth in the browser and gets one of three access levels that follow their role: member, team lead or owner. Idlefy checks the role again on every request. The Idlefy MCP server is included in Pro. The assistant cannot reach inside a machine, change billing or act beyond the person's own permissions.

How do I connect Claude Code, Claude or Cursor to Idlefy?

Add the server to your assistant with the steps for your client below.
The assistant opens an Idlefy sign-in page in your browser. Review the access level and approve it.
Ask for what you need in plain words.

claude mcp add --transport http idlefy https://mcp.idlefy.com/mcp/

Run in your terminal.

1. Open Settings → Connectors.
2. Choose Add custom connector.
3. Name it Idlefy and paste https://mcp.idlefy.com/mcp/ as the server URL.
4. If Claude asks how to register, choose automatic registration.
5. Sign in to Idlefy in the browser window that opens.

Claude desktop app or claude.ai. On Team or Enterprise, an organization owner adds it first; you then choose Connect.

{
  "mcpServers": {
    "idlefy": {
      "url": "https://mcp.idlefy.com/mcp/"
    }
  }
}

Add to ~/.cursor/mcp.json, then reload Cursor.

Server URL:  https://mcp.idlefy.com/mcp/
Transport:   Streamable HTTP
Sign-in:     standard OAuth 2.1, no API key
Discovery:   https://mcp.idlefy.com/.well-known/oauth-protected-resource

Any MCP client that supports remote servers with OAuth sign-in.

Example: start an EC2 machine from Claude Code for two hours

any MCP client · example Claude Code · example

you start the load-test box for two hours
tool list_vms
load-test-1 · stopped · c6i.2xlarge · us-east-1
ask This starts a paid machine for 2 h. Go ahead?
you yes
tool lease_vm
done load-test-1 is starting. It stops itself at 16:05.

What can an AI assistant do with Idlefy?

Three access levels follow your role in the organization. Your role and membership are checked again on every request, so if your role changes, your assistant's access changes with it.

Member idlefy:member

Lease, extend and release machines. Start or join a shared environment. Wait in line for a busy one. Create and change your own dev boxes where your organization uses Dev Boxes (AWS). Read your budgets.

lease_vm
extend_rental
release_rental
rent_workflow
join_waitlist
create_dev_box
get_budgets

Team lead idlefy:admin

Everything a member can do, plus, for their team: spend and usage, the idle audit, invitations, who may use which machine, and starting or stopping a machine.

get_cost_analytics
start_idle_audit_scan
invite_user
assign_vm_to_team
admin_start_vm

Owner idlefy:owner

Connect an AWS account or GCP project, find its machines, change roles and remove people.

get_aws_stack_setup
connect_aws_account
discover_vms
change_member_role

Not available through an assistant: Accepting the Terms, billing and plan changes, turning Dev Boxes on, creating teams, editing budgets and workflow settings.

How is this different from a generic EC2 or cloud MCP server?

When the work is done: When a person leases a machine, the lease has an end time and the machine stops by itself when it ends, so a machine leased for a run does not keep billing over the weekend.
Who may do what: The tool list follows the person’s role in the organization. A member’s assistant can lease what that member may use and nothing else.
Cloud credentials: No cloud keys go into the chat or the assistant’s config. Idlefy holds the cloud connection and the assistant signs in to Idlefy.

Idlefy only controls machines you tag for it; on AWS the IAM policy itself enforces that boundary.

Is it safe to let an AI assistant manage cloud VMs?

An assistant is a new kind of user. These are the rules the server enforces or tells it to follow.

Never on its own

The server tells every assistant never to stop, release or delete anything unless you ask.

Your yes first

The server tells the assistant to get your confirmation before anything that costs money, affects colleagues or cannot be undone. Deleting a dev box needs its exact name.

Rate limits

Costly and destructive actions are limited per person or per organization, so an assistant stuck in a loop stops early.

No keys, no shell

OAuth sign-in with a token that works only against this server. No tool runs commands inside a machine.

What does agent-ready mean?

Remote MCP server. Streamable HTTP on one address, for current and earlier protocol versions (2026-07-28 and 2025-11-25 and earlier).
Standard sign-in. OAuth 2.1 in the browser. There is no API key to create or paste.
Tools your client can reason about. Each tool is marked read-only, additive or destructive, so your client knows when to ask you first. Only the tools your role allows are listed.
Docs inside the server. Guides for leases, dev boxes, budgets, errors and safety are served as MCP resources. Tool and argument names are stable; a rename is announced a release ahead.
A machine-readable site. llms.txt and Markdown versions of these pages, for assistants that read the web.

Questions

Does Idlefy have an MCP server?

Yes. Idlefy runs a remote MCP server at https://mcp.idlefy.com/mcp/. Any AI assistant that supports remote MCP servers with OAuth sign-in can use it, including Claude Code, Claude and Cursor. The assistant signs in as one person and can only do what that person may do in Idlefy.

Which AI assistants work with the Idlefy MCP server?

Claude Code, Claude on desktop and the web, and Cursor have step-by-step setup on this page. Any other client that supports remote MCP servers over Streamable HTTP with OAuth sign-in works the same way: point it at https://mcp.idlefy.com/mcp/ and sign in to Idlefy in the browser. No API key is needed.

Can an AI assistant stop a machine someone is using?

Only an assistant acting for an owner, or for a team lead only for their own team, and the Idlefy server tells it to ask that person first. Every assistant is told never to stop, release or delete anything on its own, and Idlefy re-checks the person's role on every request, so a member's assistant cannot end a colleague's lease.

Does the assistant see my AWS or GCP credentials?

No. With the default setup, Idlefy connects to your AWS account or GCP project without static keys, and the assistant signs in to Idlefy through OAuth in your browser. It never receives a password, a cloud credential or an API key, and its access token works only against the Idlefy MCP server.

Which plan includes the Idlefy MCP server?

Pro, including the Pro trial. On the Free and Basic plans the Idlefy sign-in page explains that AI assistant access is not included and links to billing; nothing is connected and nothing is charged. On Pro, the person's role decides what an assistant can do.

Can the assistant run commands on my VM?

No. Idlefy has no access inside your machines, and no tool on the Idlefy MCP server runs a command, opens a shell or reads a file on a VM. The assistant works with leases, dev boxes, shared environments, budgets and reports, the same things you see in the Idlefy web app.

Let your assistant handle the machines

Start free, connect your cloud, then add Idlefy to your assistant on Pro.

Start free Compare plans