relay-inspect
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@relay-inspect Check console logs and take a screenshot to verify the page after edit"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Stop copying and pasting console logs, server errors and screenshots into your CLI. Relay Inspect gives your AI coding agent direct access to your browser—so it can see what you see, verify its own changes, and debug without asking you to copy and paste.
Relay Inspect is a lightweight tool for designers and engineers who want to spend more time building and less time debugging.
┌─ Chrome (CDP over WebSocket)
AI Coding Agent ←→ Relay Inspect (MCP over stdio) ─┤
└─ Dev Servers (child processes)Relay Inspect is a bridge between the Chrome DevTools Protocol, your dev server and your agent. It exposes browser state as MCP tools—console output, network requests, DOM queries and screenshots. Your agent edits code, the dev server hot reloads, and the agent verifies the result itself.
Looking for annotations? The browser annotation overlay has moved to its own package: Annoku.
Why Relay Inspect over Chrome DevTools MCP?
Google's Chrome DevTools MCP is a full browser automation tool—it clicks buttons, fills forms, runs Lighthouse audits, and scripts interactions. Relay Inspect gives agents continuous visibility while they work, so mistakes get caught in the moment, not at the end.
Gives agents eyes. Real-time access to console output, network requests, and DOM state — so your agent sees what's happening as it happens, not after it's already moved on.
Tight feedback loop. The edit → reload → verify cycle happens in a single turn. Your agent makes a change, waits for the dev server to reload, and immediately confirms it worked—without additional back-and-forth.
Focused tools, fewer wrong turns. A handful of purpose-built tools means your agent reaches for the right one rather than thrashing through a broad API surface. Less noise, more signal.
Zero overhead, zero telemetry. Connects lazily on the first tool call—no Puppeteer, no phoning home. Nothing runs until your agent needs it.
Related MCP server: @nimbus21.ai/chrome-devtools-mcp
Tools
Your agent gets access to the following tools automatically via MCP:
Browser Inspection
Tool | Description | Key Parameters |
| Execute a JavaScript expression in the browser and return the result |
|
| Return buffered console output (logs, warnings, errors) |
|
| Return captured network requests and responses |
|
| Get full request/response body for a specific network request |
|
| Query the DOM with a CSS selector and return matching elements' outer HTML |
|
| Capture a screenshot of the current page |
|
Page Control
Tool | Description | Key Parameters |
| Reload the current page (optionally bypass cache) |
|
| Wait N seconds then return new console output captured during the wait |
|
| Switch to a specific Chrome page target by ID or URL match |
|
| Navigate the current page to a new URL |
|
Server Management
Tool | Description | Key Parameters |
| Start a dev server or background process and capture its output |
|
| Read stdout/stderr output from a managed server process |
|
| Stop a running managed server process |
|
| List all managed server processes and their status | — |
Diagnostics
Tool | Description | Key Parameters |
| Check Chrome connection status and diagnose issues (does not auto-launch) | — |
Setup
Prerequisites
Node.js 20+
Chrome (or any Chromium-based browser)
Add to your MCP client
No install required — npx downloads and runs the package on first use.
Claude Code — add to .mcp.json or .claude/settings.json:
{
"mcpServers": {
"relay-inspect": {
"command": "npx",
"args": ["-y", "relay-inspect"]
}
}
}Codex CLI:
codex mcp add relay-inspect -- npx -y relay-inspectopencode — add to opencode.json:
{
"mcp": {
"relay-inspect": {
"type": "local",
"command": "npx",
"args": ["-y", "relay-inspect"]
}
}
}Chrome is auto-launched on first tool call if it isn't already running. To disable this or customize behavior, see Configuration below.
Configuration
Environment Variable | Default | Description |
|
| Chrome debugging port |
|
| Chrome debugging host |
|
| Auto-launch Chrome if not already running |
| (auto-detect) | Override Chrome/Chromium executable path |
| (none) | URL to open when Chrome is auto-launched (e.g. |
| (none) | Connect directly to a CDP WebSocket URL, skipping Chrome discovery |
|
| Max console entries to buffer |
|
| Max network requests to buffer |
|
| Max log entries per managed server |
If Chrome is already running with --remote-debugging-port, Relay Inspect will connect to it directly without launching a new instance.
Development
git clone https://github.com/samsolomon/relay-inspect.git
cd relay-inspect
npm installnpm run dev # Run with tsx (auto-recompile)
npm run build # Build with tsup
npm start # Run the built bundle
npm test # Run tests with vitestFor detailed architecture, conventions, and CDP implementation notes, see CLAUDE.md.
Available Tools
15 toolscheck_connectionA
Check Chrome DevTools connection status and diagnose issues (does not auto-launch Chrome)
Inputs: no named parameters.
connect_to_pageB
Connect to a specific Chrome page target by ID or URL pattern.
Inputs: id, waitForMs, urlPattern, url_pattern.
evaluate_jsB
Execute a JavaScript expression in the browser and return the result
Inputs: expression (required).
get_console_logsB
Return buffered console output (logs, warnings, errors) from the browser
Inputs: clear.
get_elementsA
Query the DOM with a CSS selector and return matching elements' outer HTML
Inputs: selector (required), limit.
get_network_request_detailA
Get full request and response body for a specific network request by ID
Inputs: requestId (required).
get_network_requestsC
Return captured network requests and responses from the browser
Inputs: clear, filter.
get_server_logsB
Read stdout/stderr output from a managed server process
Inputs: id (required), clear.
list_serversA
List all managed server processes and their status
Inputs: no named parameters.
reload_pageA
Reload the current page (optionally bypass cache)
Inputs: ignoreCache.
start_serverB
Start a dev server or background process and capture its output
Inputs: id (required), command (required), cwd, env, +3 more.
stop_serverA
Stop a running managed server process
Inputs: id (required).
take_screenshotB
Capture a screenshot of the current page
Inputs: format, quality.
wait_and_checkA
Wait N seconds then return new console output captured during the wait (useful after page reload)
Inputs: seconds.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 15 tools
Each tool has a clearly distinct purpose: browser connection, navigation, DOM inspection, console/network logs, screenshots, and server management. No two tools overlap in functionality, so an agent can reliably select the correct tool.
All tool names follow a consistent verb_noun snake_case pattern (e.g., evaluate_js, get_console_logs, take_screenshot). The pattern is uniform and predictable across the entire set.
With 15 tools, the set covers the browser inspection and server management domain comprehensively without being bloated. Each tool serves a necessary function, and the count feels well-scoped.
The tool set covers the full lifecycle of browser inspection (connect, navigate, evaluate, capture, logs, network) and server management (start, stop, list, logs). Any missing actions (e.g., clicking elements) can be achieved via evaluate_js, so there are no critical gaps.
Related MCP Connectors
Live browser debugging for AI assistants — DOM, console, network via MCP.
Web scraping, browser automation, screenshots and AI data extraction for AI agents via MCP.
Real Chrome for coding agents over MCP: browse, replay playbooks, human takeover, HIPAA & BAA.
A paid remote MCP for AI agent browser DevTools MCP, built to return verdicts, receipts, usage logs,
Related MCP Servers
- AlicenseNot gradedqualityDmaintenanceMCP server for browser debugging, inspection, and verification that streams console logs, network errors, and user actions into AI coding assistants.67AGPL 3.0
- AlicenseNot gradedqualityCmaintenanceLets AI coding agents control and inspect a live Chrome browser via MCP, providing Chrome DevTools capabilities for automation, debugging, and performance analysis.6 npmApache 2.0
- FlicenseNot gradedqualityDmaintenanceMCP server that connects AI agents to browser DevTools via CDP, enabling real-time access to console logs, network requests, and page state.-
- AlicenseNot gradedqualityBmaintenanceEnables AI coding agents to control and inspect a live Chrome browser via MCP, including performance tracing, network and console debugging, screenshots, and reliable Puppeteer-based automation.2,158,323 npmApache 2.0