For LLM clients (like Claude, ChatGPT, Cursor, etc.), they need to explicitly support MCP in their app or interface. For example, Claude Desktop has built-in MCP support (you can see this via the hammer icon for tools), but ChatGPT’s browser-based UI doesn’t yet allow users to connect to MCP servers.
For external tools (like Slack, WhatsApp, GitHub, etc.), it’s a bit more flexible. If a tool exposes a public API, anyone can create an MCP server for it, even without the company’s involvement. That’s why we’ve seen lots of user-created MCP servers built around existing APIs.
However, some companies, like GitHub, have chosen to create and maintain official MCP servers, which tend to be more secure, feature-complete, and easier to trust.
So to summarize
- Yes, clients must opt in to support MCP usage
- Tool makers don’t have to, but it helps! Otherwise, users may build their own unofficial tools regardless.
Do individual tools need to “opt in” or otherwise create code to use the MCP?
For LLM clients (like Claude, ChatGPT, Cursor, etc.), they need to explicitly support MCP in their app or interface. For example, Claude Desktop has built-in MCP support (you can see this via the hammer icon for tools), but ChatGPT’s browser-based UI doesn’t yet allow users to connect to MCP servers.
For external tools (like Slack, WhatsApp, GitHub, etc.), it’s a bit more flexible. If a tool exposes a public API, anyone can create an MCP server for it, even without the company’s involvement. That’s why we’ve seen lots of user-created MCP servers built around existing APIs.
However, some companies, like GitHub, have chosen to create and maintain official MCP servers, which tend to be more secure, feature-complete, and easier to trust.
So to summarize
- Yes, clients must opt in to support MCP usage
- Tool makers don’t have to, but it helps! Otherwise, users may build their own unofficial tools regardless.