ChromeBrowserMCP
A read-only bridge that lets ChatGPT inspect open Chrome tabs, exposing tools to list, read, and search tabs without accessing sensitive data.
README
Chrome Browser MCP
A local, read-only bridge that lets a private ChatGPT developer-mode app inspect the tabs already open in your desktop Google Chrome.
The bridge exposes six MCP tools:
browser_statuslist_tabsget_active_tabread_tabread_tabssearch_tabs
It does not expose cookies, local storage, session storage, saved passwords, hidden input values, arbitrary JavaScript execution, clicks, navigation, tab closing, or incognito tabs.
Proven path
The end-to-end test launches a real Chromium process with the unpacked Manifest V3 extension, starts the real native-messaging host, connects an MCP client over Streamable HTTP, opens two live pages, lists and reads them, and verifies that a password input value is not returned.
MCP client
-> http://127.0.0.1:2091/mcp
-> native host process
-> Chrome Native Messaging
-> MV3 extension
-> live Chrome tabs
Run every gate:
npm ci
npm run check
Architecture
ChatGPT developer-mode app
|
| OpenAI Secure MCP Tunnel (outbound HTTPS)
v
127.0.0.1:2091/mcp
|
| same local Node process
v
Chrome Native Messaging host
|
v
Chrome MV3 extension
|
+-- chrome.tabs
+-- chrome.scripting (isolated-world semantic extraction)
Chrome starts the native host when the extension connects. The native host starts the loopback MCP endpoint. Therefore Chrome must be open and the extension must be enabled whenever ChatGPT uses the app.
Requirements
- macOS
- Google Chrome 120+
- Node.js 20+
- A ChatGPT account with Developer Mode available
- An OpenAI Platform tunnel ID and runtime API key with Tunnels Read + Use
tunnel-client
1. Install the native host and build the extension
npm run install:mac
This installs the native-host manifest at:
~/Library/Application Support/Google/Chrome/NativeMessagingHosts/com.komaksym.chrome_browser_mcp.json
2. Load the Chrome extension
- Open
chrome://extensions. - Enable Developer mode.
- Click Load unpacked.
- Select this repository's
dist/extensiondirectory. - Confirm the extension ID is exactly:
jlpddlfiallighiohmhhkemgbhofpnha
Do not proceed if the ID differs. The native host only accepts that exact extension origin.
3. Verify the local browser chain
Keep Chrome open, then run:
npm run verify:local
A successful check prints the extension ID and the six advertised MCP tools.
4. Configure Secure MCP Tunnel
Create a tunnel and runtime API key in OpenAI Platform. Then:
export CONTROL_PLANE_API_KEY="sk-..."
./scripts/configure-tunnel.sh tunnel_0123456789abcdef0123456789abcdef
tunnel-client doctor --profile chrome-browser-mcp --explain
tunnel-client run --profile chrome-browser-mcp
The profile forwards the tunnel to:
http://127.0.0.1:2091/mcp
Keep tunnel-client run active whenever ChatGPT needs the browser tools.
5. Add it to ChatGPT
- In ChatGPT, enable Settings -> Security and login -> Developer mode.
- Open Settings -> Plugins.
- Click + to create a developer-mode app.
- Choose Tunnel as the connection type.
- Select or paste the tunnel ID.
- Use the metadata from
app-metadata.json. - Confirm ChatGPT discovers all six tools.
- In a new chat, click + -> More, select Chrome Browser, then ask:
List my open Chrome tabs.
See docs/CHATGPT_SETUP.md for exact verification and troubleshooting.
Security model
Webpage text is data, never authority. Every content result includes an explicit untrusted-content marker, and every reading tool tells the model not to follow instructions found in pages.
The extension intentionally requests access to all HTTP and HTTPS pages because it cannot summarize arbitrary open tabs otherwise. That is a powerful permission. The protection boundary is:
- the extension is loaded locally by you;
- Chrome only launches the exact allowlisted native host;
- the native host rejects any origin except the stable extension ID;
- the MCP endpoint binds only to
127.0.0.1; - the tunnel is outbound-only;
- all exposed MCP tools are read-only.
Read THREAT_MODEL.md and SECURITY_REVIEW.md before extending the tool set.
Known limitations
- One Chrome profile should load the extension at a time; two profiles can contend for port
2091. - Chrome internal pages, Chrome Web Store pages,
file://pages, and incognito tabs are not readable. - Cross-origin iframes are not traversed.
- Canvas-only applications and Chrome's built-in PDF viewer may return little semantic text.
- The extractor returns the primary document's visible text, headings, links, and description, not raw HTML. URL credentials and fragments are removed, and sensitive query parameters are redacted.
- The bridge is deliberately read-only. Browser control is a separate, higher-risk project.
Development
npm ci
npm run typecheck
npm run lint
npm test
npm run test:e2e
npm audit
The E2E test accepts:
CHROMIUM_PATH=/path/to/chromium \
CHROME_NATIVE_HOST_DIR=/path/to/native-host-dir \
npm run test:e2e
Uninstall
npm run uninstall:mac
Then remove Chrome Browser MCP from chrome://extensions and remove the ChatGPT developer-mode app/tunnel profile separately.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.