Original video in Chinese.
Key Takeaway
- By integrating AI plugins and MCP (Model Control Protocol), Raycast brings AI capabilities to the Mac and creates an “AI Mac” experience.
- As the productivity hub for Mac, Raycast can orchestrate across apps and control software through natural language interaction.
- Raycast simplifies MCP deployment and configuration, letting users easily install and use officially provided MCP servers.
- MCP lets models integrate seamlessly with external tools, improving the practicality and convenience of AI applications.
- This article highlights Raycast’s use cases in AI Chat and knowledge-base-assisted writing, and sees it as the ideal form of an AI terminal.
If you want to experience an AI Mac, just install Raycast.
Raycast is more than just the productivity hub for Mac. Once you add AI plugins, it can orchestrate across apps. I said it like this in a previous video:
An AI hub that seamlessly integrates all software, then completes every interaction in a single interface — that’s the function and experience I want from an AI PC or AI Mac.
Now, this Mac app I like best supports MCP. Look, I told GPT-4o mini: open a Safari page and go to Apple’s official website. Through MCP, the model completed the task very smoothly.
A model plus MCP is an Agent. When a terminal seamlessly integrates a model and MCP, it becomes a hardware-software integrated Agent. That’s why I especially like Raycast and pay so much attention to it.
Hello everyone, welcome back to my channel. Modestly speaking, I’m one of the few bloggers in China who can explain both the why and the how of AI clearly. What I provide is far more valuable than tutorials. Remember to follow me. If you want to connect with me, come to our newtype community. More than 1,000 friends have already joined and paid.
Back to today’s topic: Raycast MCP.
MCP has been around for so long, and there has always been a pain point: deployment and configuration are difficult for users.
As a user, you not only need to know where to find MCP servers, you also need to understand what those configuration parameters mean. Raycast does a great job here. Let me show you.
Before using MCP, install this plugin first. Bring up Raycast, type Store, and then search for MCP to see the plugin.
After installing it, search MCP and you’ll see four options: Manage, Install, Search, and Import. Don’t worry about the others — just choose Search MCP, because the official team has already prepared a bunch of commonly used MCP servers.
For example, let’s choose Playwright. Once you enter it, you’ll see that the commands and parameters are already filled in; you just need to click Install and you’re done.
Of course, Playwright is an MCP that doesn’t require environment variables. We often run into ones like Tavily, which do require an API Key, so just paste the key in and don’t worry about anything else — just install it.
This is an almost automatic deployment method. What if the official MCP list doesn’t include the one you want to use? Then you install it manually.
Type Install MCP, and you’ll see the familiar interface. Just fill in the command, parameters, and environment variables according to the instructions on GitHub. Among them, separate multiple parameters with spaces.
All installed MCPs are in Manage MCP. You can select any one of them to edit. So, once you have MCP, how do you use it in Raycast?
It’s simple: open AI Chat. Choose a model that supports AI plugins, such as GPT-4o mini. Then, in the add AI plugin option below, search for the name and attach the MCP.
So, the two clients I’m currently using that support MCP are:
One is Raycast’s AI Chat, which I use for quick Q&A. I even specifically added a shortcut for it on Stream Deck. Since the commonly used shortcuts are all taken, I assigned a rather odd and inconvenient key to AI Chat, then put it on Stream Deck, and that solved it perfectly.
The other is Cursor. Because I added a knowledge-base MCP, and Cursor has very strong document editing capabilities, whenever I need AI-assisted writing, I use Cursor.
As for ChatWise, it can only serve as a backup. You can see all my AI tools on Stream Deck now: for MCP-related tasks, I use the two I just mentioned; if MCP isn’t needed, Gemini, Grok, and NotebookLM are enough. So for clients like these, the differentiation has to be strong enough, otherwise the open rate really will be very low.
OK, that’s all for this episode. If you want to understand AI, want to become a super individual, and want to find like-minded people, come join our newtype community. See you next time!