Hyperlight
Run Python in a local hardware-isolated micro-VM that keeps state between calls
- Pricing
- Open source
- Platforms
- CLI, Linux, macOS
- Runs
- Locally
- Version
- 0.1.1
- Listing updated
- today
Similar MCP servers
About Hyperlight
Hyperlight is an open-source MCP server for developer tools. It runs locally on your machine and is started by your AI client.
Hyperlight boots a Unikraft unikernel in its own micro-VM on your machine, using KVM or MSHV on Linux, WHP on Windows and HVF on Apple silicon through the hluk CLI. It needs no account, daemon or API key, and the README says code and data stay local. The guest has no network unless you allow hosts with --net-allow, and sees none of your files unless you mount them.
Two tools do the work: run_python executes code with variables and imports persisting like a notebook, returning printed output and a trailing expression, and reset_sandbox swaps in a fresh guest from a warm snapshot, in about 60 ms on Linux with KVM. A call over the timeout (30 seconds by default) is stopped and the guest replaced. The image ships CPython 3.12 with numpy, pandas, scipy, scikit-learn, matplotlib, pillow and requests; the first start downloads roughly 130 MB and builds a snapshot, which can take seconds to minutes.
Limits are stated: Python and text only, one vCPU, pip installs lost after a reset unless passed with --package, and a call already running cannot be stopped before its timeout.
Install
Run in a terminal
claude mcp add mcp-server-hyperlight -- uvx mcp-server-hyperlightAdd to claude_desktop_config.json
{
"mcpServers": {
"mcp-server-hyperlight": {
"command": "uvx",
"args": [
"mcp-server-hyperlight"
]
}
}
}Add to ~/.cursor/mcp.json
{
"mcpServers": {
"mcp-server-hyperlight": {
"command": "uvx",
"args": [
"mcp-server-hyperlight"
]
}
}
}Add to .vscode/mcp.json
{
"servers": {
"mcp-server-hyperlight": {
"type": "stdio",
"command": "uvx",
"args": [
"mcp-server-hyperlight"
]
}
}
}Add to ~/.codeium/windsurf/mcp_config.json
{
"mcpServers": {
"mcp-server-hyperlight": {
"command": "uvx",
"args": [
"mcp-server-hyperlight"
]
}
}
}Add to your client's MCP settings (most use this shape)
{
"mcpServers": {
"mcp-server-hyperlight": {
"command": "uvx",
"args": [
"mcp-server-hyperlight"
]
}
}
}Using several servers? Build one config file with the MCP config generator.
Tools this server provides (2)
Taken from the project documentation; the running server may differ.
- run_python
- Runs Python with state kept between calls
- reset_sandbox
- Replaces the guest with a fresh one from a warm snapshot
What it does
- Micro-VM isolation
- Code runs in its own kernel via a hypervisor.
- Stateful sessions
- Variables and imports persist across calls.
- Network off by default
- Hosts must be allowed explicitly.
- Fast reset
- Fresh guest from a warm snapshot in about 60 ms.
- Preinstalled data stack
- numpy, pandas, scipy, scikit-learn and more.
- Mountable files
- Directories can be mounted read-only.
Good for
- Let an agent analyse data without touching the host
- Run untrusted generated code safely
- Prototype with pandas in a notebook-like session
- Allow only one package host for installs
Pricing
Open source
Free PyPI package
Hyperlight FAQ
Does it need an account or API key?
No, it runs locally with hluk and a hypervisor.
Can the code reach the internet?
Not unless you allow hosts with --net-allow.
What happens on timeout?
The call is stopped and the guest replaced; the model is told its variables are gone.
Is Hyperlight free?
Yes. Hyperlight is open source, so you can use and self-host it for free.
How do I add Hyperlight to Claude Code?
Run this command in your terminal: claude mcp add mcp-server-hyperlight -- uvx mcp-server-hyperlight
What are the best alternatives to Hyperlight?
The closest alternatives to Hyperlight 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.


