facebook/memlab

MemLab

memlab is an end-to-end testing and analysis framework for identifying JavaScript memory leaks and optimization opportunities.

Online Resources: [Website and Demo] | [Documentation] | [Meta Engineering Blog Post] | [AI Assistant Guide]

Features:

• Browser memory leak detection - Write test scenarios with the Puppeteer API, and memlab will automatically compare JavaScript heap snapshots, filter out memory leaks, and aggregate the results
• Object-oriented heap traversal API - Supports the creation of custom memory leak detectors, and enables programmatic analysis of JS heap snapshots taken from Chromium-based browsers, Node.js, Electron.js, and Hermes
• Memory CLI toolbox - Built-in toolbox and APIs for finding memory optimization opportunities (not necessarily just memory leaks)
• MemLens: Browser Memory Debugging Tools - Enables visualization of memory leaks and interactive memory debugging in the browser.
• Memory assertions in Node.js - Enables unit tests or running node.js programs to take a heap snapshot of their own state, perform self memory checking, or write advanced memory assertions
• MCP server for AI coding assistants - Provides an MCP server that gives AI coding assistants (Claude Code, Cursor, etc.) interactive tools to load heap snapshots, find memory leaks, and investigate optimization opportunities through natural language conversation

CLI Usage

Install the CLI

npm install -g memlab

Find Memory Leaks

To find memory leaks in Google Maps, you can create a scenario file defining how to interact with Google Maps. Let's call it test-google-maps.js:

// initial page load url: Google Maps
function url() {
  return 'https://www.google.com/maps/@37.386427,-122.0428214,11z';
}

// action where we want to detect memory leaks: click the Hotels button
async function action(page) {
  // puppeteer page API
  await page.click('text/Hotels');
}

// action where we want to go back to the step before: click clear search
async function back(page) {
  // puppeteer page API
  await page.click('[aria-label="Close"]');
}

module.exports = {action, back, url};

Now run memlab with the scenario file, memlab will interact with the web page and detect memory leaks with built-in leak detectors:

memlab run --scenario test-google-maps.js

memlab will print memory leak results showing one representative retainer trace for each cluster of leaked objects.

Retainer traces: This is the result from an example website, the retainer trace is an object reference chain from the GC root to a leaked object. The trace shows why and how a leaked object is still kept alive in memory. Breaking the reference chain means the leaked object will no longer be reachable from the GC root, and therefore can be garbage collected. By following the leak trace one step at a time, you will be able to find a reference that should be set to null (but it wasn't due to a bug).

MemLab found 46 leak(s)
--Similar leaks in this run: 4--
--Retained size of leaked objects: 8.3MB--
[Window] (native) @35847 [8.3MB]
  --20 (element)--->  [InternalNode] (native) @130981728 [8.3MB]
  --8 (element)--->  [InternalNode] (native) @130980288 [8.3MB]
  --1 (element)--->  [EventListener] (native) @131009888 [8.3MB]
  --1 (element)--->  [V8EventListener] (native) @224808192 [8.3MB]
  --1 (element)--->  [eventHandler] (closure) @168079 [8.3MB]
  --context (internal)--->  [<function scope>] (object) @181905 [8.3MB]
  --bigArray (variable)--->  [Array] (object) @182925 [8.3MB]
  --elements (internal)--->  [(object elements)] (array) @182929 [8.3MB]
...

To get a readable trace, the website under test needs to serve non-minified code (or at least minified code with readable variable, function, and property names on objects).

Alternatively, you can debug the leak by loading the heap snapshot taken by memlab (saved in $(memlab get-default-work-dir)/data/cur) in Chrome DevTool and search for the leaked object ID (@182929).

View Retainer Trace Interactively

View memory issues detected by memlab based on a single JavaScript heap snapshot taken from Chromium, Hermes, memlab, or any node.js or Electron.js program:

memlab view-heap --snapshot <PATH TO .heapsnapshot FILE>

You can optionally specify a specific heap object with the object's id: --node-id @28173 to pinpoint a specific object.

Custom leak detector: If you want to use a custom leak detector, add a leakFilter callback (doc) in the scenario file. leakFilter will be called for every unreleased heap object (node) allocated by the target interaction.

function leakFilter(node, heap) {
  // ... your leak detector logic
  // return true to mark the node as a memory leak
}

heap is the graph representation of the final JavaScript heap snapshot. For more details, view the doc site.

Heap Analysis and Investigation

View which object keeps growing in size during interaction in the previous run:

memlab analyze unbound-object

Analyze pre-fetched V8/hermes .heapsnapshot files:

memlab analyze unbound-object --snapshot-dir <DIR_OF_SNAPSHOT_FILES>

Use memlab analyze to view all built-in memory analyses. For extension, view the doc site.

View retainer trace of a particular object:

memlab trace --node-id <HEAP_OBJECT_ID>

Use memlab help to view all CLI commands.

APIs

Use the memlab npm package to start an E2E run in the browser and detect memory leaks.

const memlab = require('memlab');

const scenario = {
  // initial page load url
  url: () => 'https://www.google.com/maps/@37.386427,-122.0428214,11z',

  // action where we want to detect memory leaks
  action: async page => await page.click('text/Hotels'),

  // action where we want to go back to the step before
  back: async page => await page.click('[aria-label="Close"]'),
};
memlab.run({scenario});

MCP Server for AI Coding Assistants

The @memlab/mcp-server package provides an MCP (Model Context Protocol) server that wraps MemLab's heap analysis APIs, giving AI coding assistants (Claude Code, Cursor, Windsurf, etc.) interactive tools to explore JavaScript heap snapshots, find memory leaks, and identify optimization opportunities — all through natural language conversation.

Setup

Install globally and add to your MCP config (~/.claude.json for Claude Code, or .mcp.json for Cursor/Windsurf):

npm install -g @memlab/mcp-server

{
  "mcpServers": {
    "memlab": {
      "type": "stdio",
      "command": "memlab-mcp",
      "env": {
        "NODE_OPTIONS": "--max-old-space-size=8192"
      }
    }
  }
}

Or use npx without installing:

{
  "mcpServers": {
    "memlab": {
      "type": "stdio",
      "command": "npx",
      "args": ["@memlab/mcp-server"],
      "env": {
        "NODE_OPTIONS": "--max-old-space-size=8192"
      }
    }
  }
}

Available Tools

Once connected, the MCP server exposes tools for heap snapshot analysis including: loading snapshots, viewing summaries, finding the largest objects by retained size, looking up retainer traces, detecting detached DOM nodes, inspecting closures, searching nodes by class/property/pattern, analyzing duplicated strings, and more. See the @memlab/mcp-server README for the full tool reference and example workflows, and the MCP Investigation Skill for a structured methodology on how AI assistants can systematically investigate memory leaks using the MCP tools.

Visual Debugging for Memory Leaks in Browser

Please check out this tutorial page on how to use MemLens (a debugging utility) to visualize memory leaks in the browser for easier memory debugging.

Memory Assertions

memlab makes it possible for a unit test or running Node.js program to take a heap snapshot of its own state and write advanced memory assertions:

// save as example.test.ts
import type {IHeapSnapshot, Nullable} from '@memlab/core';
import {config, takeNodeMinimalHeap} from '@memlab/core';

class TestObject {
  public arr1 = [1, 2, 3];
  public arr2 = ['1', '2', '3'];
}

test('memory test with heap assertion', async () => {
  config.muteConsole = true; // no console output

  let obj: Nullable<TestObject> = new TestObject();
  // get a heap snapshot of the current program state
  let heap: IHeapSnapshot = await takeNodeMinimalHeap();

  // call some function that may add references to obj
  rabbitHole(obj);

  expect(heap.hasObjectWithClassName('TestObject')).toBe(true);
  obj = null;

  heap = await takeNodeMinimalHeap();
  // if rabbitHole does not have any side effect that
  // adds new references to obj, then obj can be GCed
  expect(heap.hasObjectWithClassName('TestObject')).toBe(false);
}, 30000);

For other APIs check out the API documentation.

AI Assistant Guide

The AI.md file provides structured guidance for AI coding assistants (Claude Code, Cursor, GitHub Copilot, ChatGPT, etc.) working with MemLab. It covers:

• Creating test scenarios — how to write MemLab scenario files using the IScenario interface, correct Puppeteer Page API usage, and common pitfalls to avoid
• Interpreting retainer traces — how to read the reference chain from GC root to a leaked object and identify which reference to break
• Using heap analysis plugins — the built-in analysis plugins available via memlab analyze and how to use the programmatic API
• MCP server tools — how AI assistants can use the @memlab/mcp-server to interactively load, query, and analyze heap snapshots

Development

Use node version 16 or above. To build on Windows, please use Git Bash.

First build the project as follows:

npm install
npm run build

Then keep this helper script running to ensure that local changes are picked up and compiled automatically during development:

npm run dev

NOTE: To run the memlab cli locally, make sure to prefix the memlab command with npx from within the memlab repo e.g. npx memlab

Run tests:

npm run test

License

memlab is MIT licensed, as found in the LICENSE file.

Contributing

Check our contributing guide to learn how to contribute to the project.

Code of Conduct

Check our Code of Conduct to learn more about our contributor standards and expectations.