Original video in Chinese.
Key Takeaway
- Prompt House is a tool for managing and invoking prompts. Through the MCP protocol, it lets models automatically retrieve and use prompts, chaining them together into workflows.
- I demonstrate, using the Mac version of Prompt House, how to nest multiple MCP tools and prompts inside a single prompt to automate complex tasks.
- The core value of Prompt House lies in combining MCP tools and prompts to form workflow-based automation capabilities.
- Compared with the web version, the Mac version of Prompt House offers local data storage, more powerful AI polishing features (supporting local models or custom API keys), and faster speed.
- The article emphasizes that products in the AI era should rise with the tide, conveying methodology through the product and continuously developing new features to meet user needs.
I finally finished the Mac version of Prompt House.
Building the Mac version was much more troublesome than developing the web version. It took me nearly a week and cost two hundred dollars, and I made it with the most advanced coding model available right now, Claude Opus 4.
My wife said I had two more gray hairs at my temples. I said, gray hair is OK, as long as I don’t go bald.
I won’t go into too much detail about the difficulties along the way here. I posted a rather long summary in the community before, so you can go look at it yourselves. So why did I go to so much trouble to build both the web version and the Mac version? Because I have a set of advanced prompt techniques, but there isn’t such a tool on the market that can make them happen. So I built one myself.
Let me show you, and you’ll understand its value.
I prepared two prompts in advance in Prompt House:
-
Deep Research: explicitly requires the use of two MCP tools — Sequential Thinking and Tavily — for at least three rounds of decomposition, analysis, and collection. When that is finished, another prompt, “Summary Report,” is used to complete the final output.
-
Summary Report: requires clear logic, a reasonable structure, concise expression, and so on, and each subsection must have a clear conclusion and supporting evidence.
In the client app (I’m using ChatWise here), I @prompthouse and ask it to perform deep research.
Then the model knows two things: first, it needs to use the Prompt House MCP; second, it needs to obtain the prompt for deep research.
So the model first calls the first tool to get a list of all prompts on Prompt House, finds the “Deep Research” prompt from it, and then retrieves the full content of that prompt.
Because in that prompt I explicitly require combining two MCP tools for multiple rounds of research, the model starts calling Sequential Thinking and Tavily, one for analysis and one for collection, gathering all the necessary materials.
Once all the materials are ready, it’s time to produce the final report.
Because at the end of the “Deep Research” prompt I explicitly require that if a report is to be output, another “Summary Report” prompt must be used. So the model calls the Prompt House MCP again, obtains the “Summary Report” prompt, and completes the entire report according to the detailed requirements inside it.
You see, with Prompt House connecting everything together, the model automatically runs through an entire workflow, involving the invocation of multiple MCP tools as well as multiple prompts.
There are so many MCP tools today. So when should you use which tool? How should the tools be combined? All of this can be specified clearly in a single prompt.
You also don’t need to spell out all the work in one prompt. You can completely break it down into smaller subtasks and put them into multiple prompts. Then, as needed, you can flexibly choose a few and combine them.
With Prompt House as the orchestration hub, we get a workflow based on MCP tools + Prompts, enough to automate some complex tasks.
I said before in the community that the core of all tools and all SaaS products is methodology. Through Prompt House, I fully express my methodology for prompts and share it with everyone.
The stronger the model’s ability to call tools, and the longer the context length, the greater the value this methodology and these tools can deliver. I firmly believe that in the AI era, we must build products that rise with the tide.
If you get it, feel free to log in to prompthouse.app and use it. The web version is free.
If you want to use it long term, I recommend buying the Mac version. Click Purchase in the lower-left corner of the page to buy it, for $9.98. I thought about it, and decided not to do a subscription model after all. A one-time purchase is more comfortable for everyone.
Why spend these $9.98? The Mac version has three advantages:
First, all data is stored locally, so you don’t need to worry about privacy.
Second, the “AI polishing” feature is more powerful. The web version uses my API key. To save costs, I chose GPT-4o mini. Its performance is a bit worse, but it’s dirt cheap.
On the Mac version, there are two options: you can use Ollama to call open-source models. Or you can use your own API key and choose a better model, such as GPT-4.1.
The “AI polishing” feature can not only help you polish prompt content, but also generate titles and tags. That way, you don’t have to spend energy thinking about those things. For the model, accurate titles and rich tags can significantly improve its filtering efficiency and accuracy.
Third, it’s faster and the experience is better. After all, it’s software, so it’s definitely smoother to use than a webpage. And because the MCP server is local, the connection speed is also faster.
At the moment, all the features are MVP features of Prompt House. Based on this, starting next week, I’ll be developing several more features one after another. For example, a prompt recommendation feature. I’ll keep collecting useful prompts and push them to users through the product. Users can save them into their own prompt library with a single click.
In addition, the integration between Prompt House and client apps will continue to be improved. Currently, Cursor, ChatWise, and Cherry Studio are all handled. But I found that ChatGPT can connect to the web version’s MCP, yet cannot connect to the Mac version. That issue will also be fixed next week.
If you often need to use certain prompts frequently, then Prompt House can definitely help you.
If you have a large number of prompts, then these $9.98 for the Mac version are extremely worthwhile. Using the setup I just introduced will definitely change the way you used it before.
If you’re interested in my entire development process and want to learn more AI usage tips, feel free to join newtype.
OK, that’s all for this episode. If you want to learn about AI, want to become a super individual, and want to find like-minded people, come join our community. See you next time!