Claude
Anthropic’s chat app, on the web, desktop and phone. It adds MCP servers as “connectors”.
An MCP server adds abilities, but you need an app that calls it: a chat app, a code editor or a terminal tool. These are apps whose makers say in their own docs that they support MCP, with pros and cons you can check.
MCP has two sides. The server holds the tools, such as reading a calendar or searching a shop. The app, often called an MCP client, is the program you talk to: it decides when to call those tools and shows you the result.
The same server works in any app that supports MCP, so you can change apps without rebuilding your tools. What differs is which servers each app can reach, what it costs, and who runs it.
You type in plain language. Web versions can only reach servers in the cloud, because a web page cannot start programs on your computer.
Built for writing software. Most can run servers on your computer as well as reach servers in the cloud.
Run in a command line. You add servers with a command or a config file, and they fit into scripts.
You install them on your own server, so you choose the models and keep the data, and you also do the upkeep.
The app starts the server as a small program on your machine. It can reach your files, but needs a desktop app or terminal, and often Node.js or Python.
The app connects to the server over the internet with a URL, sometimes after you sign in. It works in web and mobile apps; your requests go through the server’s owner.
Anthropic’s chat app, on the web, desktop and phone. It adds MCP servers as “connectors”.
OpenAI’s chat app. It connects to MCP servers in the cloud, which it shows as apps or connectors.
An open-source agent with a desktop app and a command line. Its add-ons (extensions) are MCP servers.
A desktop app for running language models on your own computer, which can also call MCP servers.
A desktop chat app for local and online models. Its Toolbox adds MCP servers as tools.
An open-source desktop chat app for many model providers, with MCP servers built in.
A code editor built around an assistant. It runs MCP servers on your computer and connects to cloud servers.
Microsoft’s free code editor. GitHub Copilot’s agent mode uses MCP servers.
Microsoft’s IDE for Windows. GitHub Copilot inside it uses MCP servers.
A code editor by Cognition, renamed from Windsurf to Devin Desktop. Its Cascade agent uses MCP servers.
An open-source code editor whose built-in agent uses MCP servers.
AWS’s coding app, with an editor, a command line and a web version, all able to use MCP servers.
An open-source coding agent that runs inside VS Code and other editors, and uses MCP servers.
An open-source assistant for VS Code and JetBrains, with a command-line version, that uses MCP in agent mode.
The assistant built into JetBrains IDEs such as IntelliJ IDEA and PyCharm, together with the Junie agent. Both use MCP servers.
Anthropic’s coding agent for the terminal. It runs MCP servers on your computer and connects to cloud servers.
OpenAI’s open-source coding agent for the terminal, which uses MCP servers.
Google’s open-source agent for the terminal, which uses MCP servers.
An open-source coding agent for the terminal, with a desktop app, that uses MCP servers.
An open-source terminal agent written in Common Lisp. An “mcp” cell gives it tools from MCP servers.
A terminal app with a built-in agent that uses MCP servers.
An open-source chat website you run on your own server, for many models, with MCP support.
A chat website you run on your own server, for Ollama and other model providers, that connects to MCP servers.
A launcher for Mac, Windows and iOS. Its AI chat can call MCP servers.
How we choose: the maker’s own docs say it supports MCP, it was updated in the last 6 months, and open-source apps have at least 100 GitHub stars. 25 apps, checked Nov 10, 2026. We do not rank or score them.