fleet
Let a coding agent see and use every GPU machine you can reach over SSH
- Pricing
- Open source
- Platforms
- CLI, self-hosted, Linux
- Runs
- Locally
- Version
- 0.6.2
- Listing updated
- today
Similar MCP servers
About fleet
fleet is an open-source MCP server for cloud and infrastructure. It runs locally on your machine and is started by your AI client.
Coding agents normally see only the machine they run on. fleet gives them a picture of all your machines: each GPU and what is running on it, which models are being served and where, installed software such as CUDA or ROCm and Python packages, and the folders where models and datasets live. It reaches machines with the SSH you already use and installs nothing on them.
The agent can answer questions such as what is free right now, which box has a 24 GB card, who is using the GPUs, whether Llama 3 is already served somewhere, and which rented machines are idle and costing money. Matching is by capability, such as NVIDIA or AMD, VRAM, cores and RAM, rather than by name, and the agent is told to ask which machine you mean and not to start work on a GPU someone else is using.
The README lists the safety stance: no passwords stored or requested, no keys written, secrets in GPU job command lines blanked, no telemetry and no account. It installs once on the machine you work from and registers with a long list of agents; the PyPI package agents-fleet serves MCP over stdio. A fleet command line mirrors everything the agent does.
Install
Run in a terminal
claude mcp add fleet -- uvx agents-fleetAdd to claude_desktop_config.json
{
"mcpServers": {
"fleet": {
"command": "uvx",
"args": [
"agents-fleet"
]
}
}
}Add to ~/.cursor/mcp.json
{
"mcpServers": {
"fleet": {
"command": "uvx",
"args": [
"agents-fleet"
]
}
}
}Add to .vscode/mcp.json
{
"servers": {
"fleet": {
"type": "stdio",
"command": "uvx",
"args": [
"agents-fleet"
]
}
}
}Add to ~/.codeium/windsurf/mcp_config.json
{
"mcpServers": {
"fleet": {
"command": "uvx",
"args": [
"agents-fleet"
]
}
}
}Add to your client's MCP settings (most use this shape)
{
"mcpServers": {
"fleet": {
"command": "uvx",
"args": [
"agents-fleet"
]
}
}
}Using several servers? Build one config file with the MCP config generator.
What it does
- Free-GPU lookup
- Checks every machine for idle GPUs and free VRAM.
- Job visibility
- Shows every job on every card with user, VRAM, duration and command.
- Served models
- Lists models being served and how to reach them.
- Software inventory
- Reports CUDA or ROCm, Python environments and package versions.
- SSH only
- Probes machines over SSH with nothing installed on them.
- Rental cost flags
- Flags idle paid rentals with their hourly price.
Good for
- Run training on whichever machine has a free 24 GB card
- Find out whether a model is already served on your network
- See who is using the GPUs before starting a job
- Spot idle rented GPUs that are billing you
- Run tests on the machine that runs Linux
Pricing
Open source
Free; no account or cloud service
fleet FAQ
Does it install anything on my servers?
No. The machines only need SSH; the README lists the few files it leaves in a guide.
Does it store passwords?
No. It never asks for or stores one, and the agent never sees credentials.
Does it send telemetry?
No. Apart from installing and updating itself, it talks only to your machines.
Which operating systems can I add?
Linux, macOS and Windows machines reachable by SSH.
Is fleet free?
Yes. fleet is open source, so you can use and self-host it for free.
How do I add fleet to Claude Code?
Run this command in your terminal: claude mcp add fleet -- uvx agents-fleet
What are the best alternatives to fleet?
The closest alternatives to fleet are REA, Context7, Chrome DevTools MCP, Codebase Memory and GitHub.
Where this comes from
- Description: written from the project's README.
- Install steps: from the official MCP Registry entry.


