How to add MCP Servers to Claude Desktop with Docker MCP Toolkit

What if you could turn Claude from a conversational assistant into a development partner that actual

What if you could turn Claude from a conversational assistant into a development partner that actually things—safely, securely, and without touching your local machine? If you’ve been exploring Claude Desktop and wondering how to connect it with real developer tools, is the missing piece you’ve been looking for. Here’s the reality: Today’s AI assistants are brilliant at reasoning and explaining concepts. But when it comes to on that knowledge? They’re stuck. They can’t deploy containers, manage repos, or analyze data without trusted tools to bridge that gap. That’s where Docker MCP Toolkit comes in. Think of it this way: Together, they transform Claude from something that just talks to something that —all through Docker’s trusted, security-hardened workflow. In this guide, I’ll walk you through setting up Claude Desktop as an MCP client, connecting it to MCP servers running in Docker Desktop, and automating real developer tasks securely and reproducibly.  Imagine taking a screenshot of any app UI—a Stripe checkout page, a dashboard component, anything—and watching Claude recreate it as a fully-tested, production-ready React component, complete with Jest tests, GitHub repo, and deployment. That’s exactly what we’ll build together Before we dive into the setup, let’s clarify what MCP actually is. (MCP) is the standardized way AI agents like Claude connect to tools, APIs, and services. It’s what lets Claude go beyond conversation and perform real-world actions—like deploying containers, analyzing datasets, or managing GitHub repositories. MCP is the bridge between Claude’s reasoning and your developer stack. And Docker? Docker provides the guardrails that make it safe. I’ve been working with AI tools for a while now, and this Docker MCP setup is one of the most impressive things I’ve seen. Docker MCP Toolkit bridges Claude’s intelligence with Docker’s trusted developer workflow. Docker MCP Toolkit doesn’t just connect Claude to your tools—it creates a where Claude can work without any risk to your local machine. Every action happens in an isolated container that gets torn down when you’re done. No mess, no security risks, complete reproducibility. Once connected, Claude Desktop can tap into Docker’s curated to automate developer tasks that used to eat up hours of your day. Here’s what that looks like in practice: Each of these MCP servers can be enabled directly through the MCP Toolkit in Docker Desktop. The setup is straightforward, the execution is secure, and the results are reproducible every single time. Before you begin, make sure you have: Head over to and download if you haven’t already. The installation is pretty straightforward – just follow the prompts. Once it’s installed, sign in with your Anthropic account. Open Docker Desktop and enable the MCP Toolkit from the Settings menu. Click “MCP Toolkit” in the left sidebar of Docker Desktop and click on “Connect” under Claude Desktop. Restart Claude Desktop and verify that the MCP_DOCKER connection appears under the Claude Desktop menu. You can view the Claude configuration file claude_desktop_config.json accessible via Claude > Settings > Developers > Edit Config. Click “Edit Config” and open it via Visual Studio Code or your preferred IDE. When you connect Claude Desktop to Docker MCP Toolkit, this configuration gets added to Claude Desktop’s claude_desktop_config.json file: Here’s what each part does: When Claude Desktop needs to access MCP tools, it executes docker mcp gateway run, which: In simple terms: Think of it like this: This one configuration line is what makes all the magic happen—it’s the handshake that lets Claude safely access and execute all those containerized developer tools. Where to find this file: When you click “Connect” in Docker Desktop’s MCP Client settings, this configuration gets added automatically—you don’t need to edit it manually. Now that you’ve connected Claude Desktop to , let’s see it in action with a practical example. What if you could take a screenshot of any app, and have Claude recreate it with working code AND full test coverage in minutes? Right now, the typical workflow looks like this: It’s slow, manual, and riddled with back-and-forth communication gaps. Claude Desktop + automates all three steps. You provide a screenshot, and Claude handles the rest—analysis, code generation, testing, debugging, and deployment. A functional React component from a screenshot, complete with: Note: This is a code foundation, not production deployment. Does NOT include: CI/CD, live deployment, Playwright tests, or a11y validation. For this demonstration, we’ll use five MCP servers: Phase 1: Screenshot to Component – You provide a screenshot, could be a Stripe checkout page, a dashboard card, whatever UI you need. – Claude analyzes the layout, identifies components, measures spacing, captures colors, and generates a React/Next.js component that matches the design. Phase 2: Generate Test Suite – Jest unit tests for component logic (41 tests covering rendering, interactions, filters and state management). Phase 3: Run & Debug with Node.js Sandbox – The Node.js Sandbox MCP spins up an isolated container, installs all dependencies safely, and runs your test suite. – Everything happens in a sandboxes environment; nothing touches your local machine. – If tests fail? Sequential Thinking kicks in. Claude iterates, fixes the code, spins up a container, and runs the tests again. It repeats this loop until everything passes. Phase 4: Deploy to GitHub Once tests are green: – Claude creates a GitHub repo with proper structure – Pushes the code What used to take 4 hours now takes 10 minutes. Assuming that Docker MCP Toolkit is already configured with Claude Desktop, follow the below steps to configure and add MCP servers.  Click “MCP Toolkit” and select “Catalog” to search for the following MCP servers and add them one by one. Note:  The Node.js Sandbox MCP server implements Docker-out-of-Docker (DooD) pattern by mounting /var/run/docker.sock, giving the sandbox container full Docker daemon API access. This allows it to spawn ephemeral sibling containers for code execution – when Claude requests JavaScript execution, the sandbox container makes Docker API calls to create temporary Node.js containers (node:lts-slim, Playwright, etc.) with resource limits (512MB RAM, 0.75 CPU cores), executes the code in isolation, and auto-removes the container. The Docker socket mount is a privilege escalation vector (effectively granting root-level host access) but enables dynamic container orchestration without the overhead of true Docker-in-Docker. Files are persisted via volume mount to ~/Desktop/sandbox-output, mapped to /root inside execution containers. This architecture trades security for flexibility – acceptable for local development but requires Docker Scout vulnerability scanning and careful consideration for production use. Hence, you’ll require a separate entry for Sandbox MCP server in  the Claude configuration file. You’ll see how to configure later in this article. The gives Claude Desktop the ability to add powerful web scraping and search capabilities. To setup a Firecrawl MCP server, you’ll need Firecrawl API key via . Create a new account if you’re visiting the portal for the first time. Click “API Keys’ on the left sidebar to get the new API keys created. The Firecrawl MCP server should now appear under “My Servers” in Docker MCP Toolkit. The enables Claude Desktop to create issues, PRs, and manage repositories on your behalf. No manual token creation. Authorization happens through GitHub’s secure OAuth flow with automatic token refresh. If you prefer manual control or need specific scopes: The gives Claude Desktop the ability for dynamic and reflective problem-solving through thought sequences. Adding the Sequential Thinking MCP server is straightforward –  it doesn’t require any API key. Just search for Sequential Thinking in the Catalog and get it to your MCP server list. The Sequential Thinking MCP MCP server should now appear under “My Servers” in Docker MCP Toolkit. The Sandbox MCP enables Claude Desktop to spin up disposable Docker containers to execute arbitrary JavaScript. To get it added to the Claude Desktop, replace the contents of the Claude Desktop configuration file with the following JSON structure. This configuration tells Claude Desktop to start the Sandbox with access to specific directories: Before you restart your Claude Desktop, make sure that the /Users/YOUR_USERNAME/Desktop/sandout-output directory exists on your local system and this directory is made available to containers via Docker Desktop > Settings > Resources > File Sharing. The enables Claude Desktop to access the latest and up-to-date code documentation for LLMs and AI code editors. Adding Context7 MCP server is straightforward. It doesn’t require any API key. Just search for Context7 in the Catalog and get it added to the MCP server lists. The Context7 MCP server should now appear under “My Servers” in Docker MCP Toolkit Once you have added all the MCP servers, click “Disconnect” and “Connect” so as to see the various MCP tools under MCP_DOCKER. That’s it. It’s time to start interacting with your MCP servers and tools. Ensure that you the screenshot Here’s how Claude Desktop responds. Within minutes, Claude will:  Example Result: You should be able to verify if Sandbox container is running or not using Docker Dashboard. The logs show the MCP (Model Context Protocol) server successfully initializing inside a Docker container. The container starts a scavenger process that will automatically clean up after 1 hour of inactivity (checking every 60 seconds). It then performs a handshake announcing it supports MCP Protocol 2025-06-18 and server version 1.3.0. The server registers 7 tools for JavaScript execution:  All tools that execute JavaScript require ESModules syntax and use the ./files directory for file persistence. The server is fully operational and ready to execute code!  You’ve just witnessed how Docker MCP Toolkit transforms Claude Desktop from a chat assistant into a genuine development partner. What used to require hours of manual work—analyzing designs, writing code, creating tests, debugging failures, setting up repos—now happens in minutes with a single prompt. This is the new paradigm for AI-assisted development. You’re not just using AI anymore. You’re with it in a way that’s secure, reproducible, and production-ready. Ready to try it? to get started with MCP Toolkit ( ).