Meet litmus-lite: chaos testing for developers. Ask for it in Cursor or run it from a terminal.Chaos testing for developers. Just ask.
litmus-lite injects a failure into your service, judges the run against hypotheses you wrote, and writes one HTML report. Ask for it in Cursor or run it from a terminal. One binary, scenarios as files, a verdict you can trust.
- One binary
- No cluster
- No account
- Agent optional
- Writes the scenario as a *.chaos.yaml file next to your service
- Validates it and runs it through litmus-lite's MCP tools
- Answers from the hypotheses: which held, which did not, how long recovery took
Why litmus-lite?
Know whether your service survives a slow dependency or a failing route before you merge. Describe the failure in plain words or in a small file, and get a clear verdict back.
A test is a file
Scenarios are *.chaos.yaml files that live beside the service they test. Review them, commit them, run them in CI. Nothing exists only in a chat.
Agent-native, not agent-mandatory
Cursor, or any MCP client, can write and run scenarios for you. Every MCP tool is a CLI command you can type, so the agent is optional the second time.
Hypotheses, not thresholds
A run says which of your statements held: errors under 5%, healthy again within 10 s. No data is a failure, and a service that never came back fails as not recovered.
Safe by default
Loopback targets only, PIDs you name, --yes for anything destructive, and every fault rolled back on exit or Ctrl-C. An agent cannot grant itself those permissions.
One static binary
Linux, macOS and Windows, no runtime and no daemon. No telemetry: by default nothing leaves your machine.
A report you can read
One self-contained HTML file with probe latency and errors across baseline, fault and recovery, plus JSON for tools and JUnit for CI.
Tell me what should break and what must stay true. I will write the test, run it and report back.
Built for developers and their agents
It fits the way you already work
- Scenarios are plain files in your repository: review them, commit them, change a number and run again
- Runs in a terminal or in CI, and the exit code is the verdict: 0 every hypothesis held, 3 one did not
- One binary and nothing to stand up: no cluster, no control plane, no account
- A self-contained HTML report, JSON for tools and JUnit for CI
It speaks MCP out of the box
- litmus-lite init connects Cursor in one command and writes the rules the agent follows. Claude Code and any other MCP client work too
- MCP tools to create, validate and run a scenario, then diagnose a failure and compare reports
- Every tool is a CLI command you can type yourself, so nothing lives only in the agent
- The agent answers in terms of your hypotheses, and cannot grant itself destructive permissions
litmus-lite is part of the LitmusChaos project. Read about working with an agent.
Quick start
From download to your first verdict in a few minutes. Nothing leaves your machine.
- 1
Install
Download the binary for Linux, macOS or Windows from the releases page.
$ litmus-lite versionInstall guide → - 2
Connect Cursor
Adds the MCP server and the rules that tell the agent how to behave. Any other MCP client works too.
$ litmus-lite initWorking with an agent → - 3
Ask for a failure
Describe what should break and what must stay true. Prefer a terminal? Run the same file with litmus-lite run.
Add 800 ms of latency to the checkout API for 12 seconds and tell me whether it copes.
Full 5-minute quick start →
Find your way around the docs
Everything lives on the litmus-lite documentation site. Jump straight to what you need.
Break something on purpose, before production does.
litmus-lite is open source (Apache-2.0). Install it, ask your agent for a failure, and see how your service copes. Pip will be waiting.
Read the launch post →
Copyright © 2026 LitmusChaos Authors. All rights reserved.
The Linux Foundation has registered trademarks and uses trademarks. For a list of trademarks of The Linux Foundation, please see our Trademark Usage page.
Originally created by