kernel
Open-source MCP server for secure, low-latency cloud-browser automation on Kernel.
claude mcp add --transport stdio kernel-kernel-mcp-server node path/to/server.js
How to use
Kernel MCP Server provides AI assistants with secure access to Kernel tools and browser automation through the MCP protocol. The server acts as a remote MCP endpoint that authenticates users via Clerk OAuth 2.0 and exposes Kernel capabilities such as deploying Kernel apps in the cloud, launching and controlling headless Chromium sessions for web automation, monitoring deployments and invocations, searching Kernel documentation, and running Playwright code against live browsers with automatic video replays. Clients that support MCP can connect to the streamable HTTP endpoint at https://mcp.onkernel.com/mcp or use a stdio-based transport via mcp-remote for environments without MCP support. The CLI and various integrations (Cursor, Claude, VS Code, Windsurf, Goose, OpenCode, etc.) can be configured to point at this remote MCP server to enable Kernel tools inside their workflows.
How to install
Prerequisites:
- Node.js 18+ installed on your machine or server
- npm (comes with Node.js) or yarn
- Access to the Kernel MCP Server endpoint (https://mcp.onkernel.com/mcp) and Clerk OAuth setup if you plan to run locally
Install and run locally:
-
Ensure Node.js is installed (version 18+):
node -v npm -v
-
Clone or download the Kernel MCP Server project and install dependencies:
git clone https://github.com/onkernel/kernel-kernel-mcp-server.git cd kernel-kernel-mcp-server npm install
-
Configure environment (example):
-
Set up Clerk OAuth credentials or use the hosted service. Depending on the deployment, you may need environment variables like: CLERK_FRONTEND_API={your-front-end-api} CLERK_SECRET_KEY={your-secret-key}
-
If the server expects specific Kernel integration settings, provide them in a .env file or via your hosting platform's config.
-
-
Start the MCP server:
node path/to/server.js
-
Verify:
Visit https://mcp.onkernel.com/mcp (or the URL exposed by your deployment) to ensure the MCP endpoint is reachable. Use a client configured to point at this endpoint (e.g., mcp-remote, Cursor, Claude, VS Code) to test a connection.
Additional notes
Tips and caveats:
- This server uses OAuth 2.0 via Clerk for secure access to Kernel resources; ensure Clerk credentials and OAuth flow are correctly configured for local or hosted deployments.
- When using clients that support remote MCP, prefer the streamable HTTP endpoint for reliability. If your client lacks remote MCP support, you can fall back to stdio via mcp-remote.
- For tooling integration (Cursor, Claude, VS Code, Windsurf, Goose, OpenCode, etc.), follow each tool’s specific instructions to add an MCP server, pointing to https://mcp.onkernel.com/mcp.
- If you encounter authentication or connectivity issues, check network access to the MCP endpoint and verify that the OAuth redirect URLs and client registrations are correctly configured in Clerk.
- The Kernel tools exposed via MCP include deployment management, browser automation with Playwright, documentation search, and video replays; each tool typically maps to a manage_* function with standardized action parameters.
Related MCP Servers
flyto-core
The open-source execution engine for AI agents. 412 modules, MCP-native, triggers, queue, versioning, metering.
mcp-agent
Lightweight, focused utilities to manage connections and execute MCP tools with minimal integration effort. Use it to directly call tools or build simple agents within your current architecture.
puppeteer
Self-hosted Puppeteer MCP server with remote SSE access, API key authentication, and Docker deployment. Complete tool suite for browser automation via Model Context Protocol.
mcp-json-yaml-toml
A structured data reader and writer like 'jq' and 'yq' for AI Agents
Derived-WMD
The Agentic Generative UI Platform: Transform natural language into production-ready React apps in seconds. Featuring autonomous Tambo agents, Model Context Protocol (MCP) for codebase grounding, and secure E2B sandboxed execution.
VegaMCP
VegaMCP v6.0 - Protocol Supremacy Edition. AI Agent Swarm Platform with 60+ tools, A2A protocol, zero-trust identity, MCP gateway, async tasks, OAuth 2.1, agent graphs, multimodal embeddings.