MCP Server

Overview

The Limelight MCP Server connects your running React or React Native app to AI coding assistants like Cursor, Claude Code, and any MCP-compatible editor. Instead of copying logs into ChatGPT or hoping your AI can guess what’s wrong from source code alone, Limelight streams live runtime data — renders, state changes, network requests, and console logs — directly into your editor’s AI. Ask your AI “why is my app slow?” and it answers with real data, not guesses.

The MCP server runs locally on your machine. No data leaves your system.

Quickstart

Add the MCP server to your editor

Claude Code
Cursor
Other MCP Clients

claude mcp add limelight-mcp npx limelight-mcp

Add to your MCP settings:

{
  "mcpServers": {
    "limelight": {
      "command": "npx",
      "args": ["limelight-mcp"]
    }
  }
}

Any client that supports stdio MCP servers works. Use the command:

npx limelight-mcp

Install the SDK in your app

npm install @getlimelight/sdk

Initialize with the MCP target:

import { Limelight } from "@getlimelight/sdk";

Limelight.connect({
  target: "mcp",
});

Add Zustand or Redux stores to capture state changes:

Limelight.connect({
  target: "mcp",
  stores: { authStore: useAuthStore, cartStore: useCartStore },
});

Start your app and ask your AI

With your app running and your AI editor open, try:

“My app feels slow. Do you see any issues?”
“Why is my search showing wrong results?”
“Which components are re-rendering the most?”

Your AI will call Limelight’s tools automatically to inspect your app’s runtime state and give you answers backed by real data.

What your AI can see

Once connected, your AI assistant has access to everything happening in your running app:

Renders

Which components are rendering, how often, how expensive, and why. Detects render loops, unnecessary re-renders, and unstable props.

State

Zustand and Redux store contents, recent changes, and diffs. See exactly how state evolved over time.

Network

Every request and response with timing, status, headers, and bodies. Detects race conditions, waterfalls, and failed requests.

Console

All console output with levels, timestamps, and stack traces. Filtered and searchable.

Tools reference

The MCP server exposes 10 tools that your AI calls automatically based on your questions.

Diagnostics

limelight_get_session_overview

High-level snapshot of your running app. Event counts, errors, suspicious components, and detected patterns. This is usually the first tool your AI calls.Returns: total events by type, error/warning counts, top rendered components, suspicious items, and session metadata.

limelight_find_issues

Proactive scan across all captured events. Runs Limelight’s correlation engine and Debug IR pipeline on anything that looks problematic.Detects: unnecessary re-renders, unstable props, render cascades, race conditions, N+1 queries, state thrashing, and more.

Parameter	Type	Default	Description
`severity`	`"info" \| "warning" \| "critical"`	all	Minimum severity
`limit`	number	5	Max issues to return
`verbose`	boolean	false	Include full analysis detail

limelight_investigate_error

Full root cause analysis on an error. Runs the Debug IR pipeline to produce a causal chain, state deltas, violations, and suggested fixes.

Parameter	Type	Default	Description
`error_id`	string	—	Specific event ID
`error_pattern`	string	—	Match against error messages
`scope`	`"most_recent" \| "all"`	`"most_recent"`	Which errors to analyze

Querying

limelight_query_network

Filter and search captured network requests.

Parameter	Type	Default	Description
`url_pattern`	string	—	URL substring or pattern
`method`	string	—	HTTP method filter
`status_range`	`{ min, max }`	—	Status code range
`min_duration_ms`	number	—	Slow request threshold
`include_bodies`	boolean	false	Include request/response bodies
`limit`	number	10	Max results

limelight_query_logs

Filter and search console events.

Parameter	Type	Default	Description
`level`	`"error" \| "warn" \| "log" \| "info" \| "debug"`	—	Log level filter
`message_pattern`	string	—	Search within messages
`include_stack_traces`	boolean	auto	Include stack traces
`limit`	number	50	Max results

limelight_get_timeline

Chronological view of all events within a time range.

Parameter	Type	Default	Description
`last_n_seconds`	number	30	Time window
`event_types`	array	all	Filter by type
`min_severity`	`"info" \| "warning" \| "error"`	—	Minimum importance

Deep dives

limelight_investigate_component

Full analysis of a React component — render history, props driving re-renders, and correlated state/network activity.

Parameter	Type	Description
`component_name`	string	Component to investigate

limelight_get_render_profile

Component render performance profiling. Shows render counts, costs, velocity, cause breakdown, and suspicious flags.

Parameter	Type	Default	Description
`component_name`	string	—	Filter to one component
`suspicious_only`	boolean	false	Only flagged components
`sort_by`	`"render_count" \| "render_cost" \| "velocity"`	`"render_cost"`	Sort order
`limit`	number	10	Max results

limelight_get_state_snapshot

Current state store contents and recent change history.

Parameter	Type	Default	Description
`store_id`	string	—	Specific store
`path`	string	—	Dot-notation path into state
`include_history`	boolean	false	Include recent changes
`history_limit`	number	10	Number of recent changes

limelight_correlate_event

Find everything related to a specific event using Limelight’s correlation engine.

Parameter	Type	Default	Description
`event_id`	string	—	Event to correlate
`max_depth`	number	2	Hops in the correlation graph

Configuration

The MCP server accepts CLI arguments for customization:

npx limelight-mcp --port 9229 --max-events 10000 --verbose

Flag	Default	Description
`--port`	9229	WebSocket port for SDK connection
`--max-events`	10000	Maximum events stored in memory
`--verbose`	false	Enable verbose logging

How it works

Your App (with SDK) → WebSocket → Limelight MCP Server → stdio → AI Editor

The Limelight SDK captures runtime events in your app
Events stream to the MCP server over a local WebSocket connection
The MCP server runs correlation and analysis on the events
Your AI assistant calls Limelight’s tools via the MCP protocol
Responses include structured, pre-analyzed debugging context — not raw logs

All data stays on your machine. The MCP server runs locally and communicates with your editor over stdio.

The MCP server stores events in memory. Data resets when the server restarts. Maximum capacity is configurable with --max-events.

Supported frameworks

Framework	Status
React Native	Supported
React (web)	Supported
Next.js	Supported
Node	Supported

Supported state libraries

Library	Status
Zustand	Supported
Redux	Supported
Jotai	Coming soon
MobX	Planned

Getting Started

Features

Limelight MCP

MCP Server

Overview

Quickstart

What your AI can see

Renders

State

Network

Console

Tools reference

Diagnostics

Querying

Deep dives

Configuration

How it works

Supported frameworks

Supported state libraries

Getting Started

Features

Limelight MCP

​Overview

​Quickstart

​What your AI can see

Renders

State

Network

Console

​Tools reference

​Diagnostics

​Querying

​Deep dives

​Configuration

​How it works

​Supported frameworks

​Supported state libraries

Overview

Quickstart

What your AI can see

Tools reference

Diagnostics

Querying

Deep dives

Configuration

How it works

Supported frameworks

Supported state libraries