Let  AI  see  what  your  code  actually  does  at  runtime

One SDK. Full-stack runtime context delivered via MCP to Cursor, Claude Code, and Copilot.

ClaudeLimelight MCP
$Why is search showing stale results?

Three steps to runtime debugging. Add the SDK, and your AI assistant sees every network request, state change, and re-render automatically.

FIG 0.1

Add the SDK to your project

One import, one function call. Works with React, React Native, Next.js, and Node.js. No wrappers, no config files.

FIG 0.2

Limelight captures runtime context

Network requests, state changes, component renders, console logs — captured and correlated across your full stack automatically.

FIG 0.3

Your AI debugs with real data

Runtime context flows to Cursor, Claude Code, or Copilot via MCP. Your AI coding assistant debugs with actual runtime data, not guesses from source code.

What your AI actually sees

Limelight doesn't dump raw logs into your AI. It delivers a correlated causal chain — every request, state change, and re-render linked by timing and causality, structured for your AI to actually understand.

2.0 Try Limelight →

10:23:14
.0
.2
.4
.6
:15
.0
.2
.4
.6
:16
.0
.2
.4
.6
:17
.0
.2
.4
.6
:18
.0
.2
.4
.6
:19
.0
.2
.4
.6
Race condition: stale search
N+1 query storm
Render storm

Runtime issues caught and explained before you notice them

Limelight's correlation engine continuously analyzes your app's runtime — detecting patterns like N+1 queries, retry storms, and render loops, linking events across frontend, backend, and state boundaries. Your AI coding assistant gets pre-analyzed context, not a pile of logs to sift through.

3.0 See how →

AI-powered runtime debugging

Ask your AI assistant about any bug. Limelight provides the runtime evidence — network requests, state mutations, component renders — so your AI gives a specific diagnosis instead of a generic checklist.

Why did this request fail?

Cross-boundary event correlation

See how events connect across your stack. Limelight links a user action to the API call it triggered, the state change that followed, and the component re-render that resulted.

Network request

POST /api/checkout

Store change

UNAUTHENTICATED

Console error

Token refresh failed

Pattern detection catches bugs before users do

Limelight automatically detects N+1 query storms, render loops, race conditions, retry storms, and stale closures. Issues surface in your editor before users report them.

APP
Repeated warning: deprecated API usage(10)
1 affected log
APP
Unhandled error in useEffect hook(10)
3 affected logs
APP
Component rendered 50× in 2 seconds(10)
2 affected logs
APP
Query returned 3× more data than usual(10)
1 affected request

Performance insights at a glance

Limelight evaluates query complexity, detects over-fetching, and flags expensive operations. Know which API calls and renders are slowing your app without manual profiling.

Query Health
High

Score: 297

This query returns significantly more data than usual,
which may impact performance.

query GetUserProfile($userId: ID!) {
  user(id: $userId) {
    __typename
    ...UserProfileFieldsFragment
  }
}

Setup in under 60 seconds

No configuration files. No environment variables. No build plugins. One line of code and an MCP server config. That's it.

Install the package

Add the Limelight SDK to your React Native project. One package, zero peer dependencies.

npm install @getlimelight/sdk

Add two lines of code

Import Limelight and call connect(). No wrappers, no providers, no config files. Runtime capture starts immediately.

import { Limelight } from"@getlimelight/sdk"

Limelight.connect()

Add the MCP server to your editor

Connect your AI coding assistant to Limelight's MCP server so it can access your app's runtime context while debugging.

claude mcp add limelight-mcp npx limelight-mcp

Frequently Asked Questions

Everything you need to know about Limelight, the MCP server, and runtime debugging for AI coding assistants

Limelight is a runtime debugging tool built for developers who use AI coding assistants like Cursor, Claude Code, and GitHub Copilot. It captures your app's full runtime context — network requests, state changes, component renders, and console logs — correlates them into structured causal chains, and delivers that context to your AI assistant via MCP. Instead of guessing from source code, your AI sees what actually happened at runtime.
Limelight's MCP (Model Context Protocol) server connects your AI coding assistant to real-time runtime data. When you ask Cursor, Claude Code, or Copilot a debugging question, the MCP server provides correlated runtime evidence — out-of-order API responses, state overwrites, unnecessary re-renders — so your AI gives you a specific root cause and fix instead of a generic checklist. No more copy-pasting logs or screenshots.
Limelight works with any MCP-compatible AI coding assistant, including Cursor, Claude Code, GitHub Copilot, and Windsurf. Setup takes under a minute — add the MCP server config to your editor and connect the Limelight SDK in your app with two lines of code.
Debug IR (Intermediate Representation) is Limelight's structured output format optimized for AI consumption. Raw runtime events are correlated by timing, causality, and component relationships into causal chains. Patterns like N+1 queries, render loops, retry storms, and race conditions are detected automatically. The output is structured and token-efficient, so your AI coding assistant gets pre-analyzed debugging context, not raw logs.
Chrome DevTools MCP exposes raw browser data (DOM, console, network) to your AI, but it has no understanding of your application. It can't track React state changes, correlate a network request to the component render it caused, or detect patterns like N+1 queries and race conditions. It's also web-only. Limelight captures application-level runtime data across React, React Native, Next.js, and Node.js, then runs a correlation engine that links events into causal chains and automatically detects over 20 bug patterns. Your AI coding assistant gets a structured diagnosis, not a pile of raw logs.
Limelight supports React, React Native (including Expo and New Architecture), Next.js, and Node.js. The SDK is lightweight, open-source, and available on npm as @getlimelight/sdk. Setup takes two lines of code with zero configuration — no wrappers, no providers, no build plugins.
No. Limelight runs entirely on your local machine. Runtime data flows from your app to the Limelight desktop app and MCP server over localhost. The Debug IR sent to your AI coding assistant is token-efficient and automatically strips personal data, so no PII or raw application data leaves your machine.
Limelight automatically detects over 20 runtime patterns including N+1 query storms, render loops, race conditions (like stale search results from out-of-order API responses), retry storms, stale closures, unnecessary re-renders, and over-fetching. Each detected issue includes a correlated causal chain showing exactly which network request, state change, or render triggered the problem.
Yes, Limelight is free to use.

Give your AI the runtime context it's missing.