Kernel

You can access Kernel’s MCP Server to securely connect AI assistants to Kernel tools and browser automation capabilities. Use the remote HTTP endpoint for quick setup or run the server locally for development and testing. The server supports cloud deployments, headless browser control, app deployments, and documentation search, all through a unified MCP interface.

How to use

Connect using either the remote HTTP endpoint or a local stdio transport, then start issuing MCP actions from your AI assistant or client. The primary remote endpoint is https://mcp.onkernel.com/mcp. You can use this to enable your assistant to deploy Kernel apps, launch and automate headless browsers with Playwright, search Kernel documentation, and capture video replays of browser sessions.

How to install

Prerequisites you need before installing and running the MCP server locally or configuring a client: Node.js (or Bun) for running the local server, and an MCP-compatible client to connect to the server.

Quick start using the Kernel CLI (recommended for rapid setup):

# Install the CLI
brew install onkernel/tap/kernel
# or: npm install -g @onkernel/cli

# Install MCP for your tool
kernel mcp install --target <target>

Supported targets include Cursor, Claude Desktop, Claude Code, VS Code, Windsurf, Zed, Goose. The CLI automatically locates your tool’s config file and adds the Kernel MCP server configuration.

Additional notes and configuration

If you prefer not to install anything locally, you can use the hosted MCP server via the remote HTTP endpoint: https://mcp.onkernel.com/mcp. If you need to operate without remote MCP support, you can use a stdio transport with a local process that runs the MCP client. The server uses OAuth 2.0 authentication via Clerk to control access to your Kernel resources.

Security and access

Security is enforced via OAuth 2.0 authentication through Clerk. Ensure your client is configured to obtain and present valid access tokens when connecting to the server. For organization-wide usage, configure connectors or environments per your provider’s guidance to authorize each user appropriately.

Examples of common workflows

Invoke actions in your Kernel apps, manage browser sessions, generate Playwright code on the fly, and retrieve results with automatic video replay links for debugging or auditing. Your assistant can search Kernel docs, inject context, and monitor deployments or invocations through the MCP interface.

Troubleshooting

If you encounter issues when starting or using the MCP server, confirm you have a recent Node.js or Bun runtime, verify network access to https://mcp.onkernel.com/mcp, and check that your OAuth tokens are valid. If problems persist, re-authenticate your session and retry connecting to the MCP server.

Examples

Human: Run my web-scraper app to fetch data from reddit.com\nAssistant: I’ll execute your web-scraper action in the cloud and return the results.

Available tools

create_browser

Launch a new browser session with options such as headless mode, stealth, timeout, and user profile.

get_browser

Retrieve information about a specific browser session.

list_browsers

List all active browser sessions managed by Kernel MCP Server.

delete_browser

Terminate a browser session and clean up associated resources.

execute_playwright_code

Run Playwright/TypeScript code in a fresh browser session with automatic video replay and cleanup.

take_screenshot

Capture a screenshot of the current page, with optional region specification.

setup_profile

Create or update browser profiles through a guided setup process.

list_profiles

List all available browser profiles.

delete_profile

Delete a browser profile permanently.

list_apps

List apps in your Kernel organization with optional filtering.

invoke_action

Execute actions in Kernel apps.

get_deployment

Get deployment status and logs for a given deployment.

list_deployments

List all deployments with optional filtering.

get_invocation

Retrieve details about a specific action invocation.

search_docs

Search the Kernel platform documentation and guides.