In recent years, the field of Artificial Intelligence has witnessed rapid technological advancements, giving rise to a plethora of new terms such as MoE, Reinforcement Learning, Agents, computer-use, and A2A. For ordinary users lacking a technical background, these terms and technological concepts can be overwhelming, leading to a significant cognitive burden. Consequently, their interaction with AI is often limited to simple question-and-answer exchanges within chat boxes.
MCP, or Model Context Protocol, is one of these technical concepts. Over the past year, AI agents have rapidly evolved, and MCP protocols have emerged as a critical underlying capability supporting complex task automation. However, the current MCP revolution still seems to be the exclusive domain of developers, with obscure protocol documentation, complex tool registration, and high barriers to personalized configuration. As a result, most ordinary users can only observe from a distance and find it difficult to truly get hands-on experience.
However, this situation is changing. On April 23rd, Nano AI, a subsidiary of 360, announced the launch of the ‘MCP Toolbox’ designed for individual users. This product is tailored for ordinary users without a technical background, enabling everyone to master cutting-edge AI usage with minimal learning costs.
This product not only fully supports the MCP protocol but can also run agent tasks based on various large model infrastructures. Additionally, it boasts powerful capabilities such as automatically invoking external tools, accessing AI knowledge bases, and supporting user-defined task flows. Crucially, the operational threshold has been significantly lowered, requiring no coding skills, and can be used simply by opening a chat box.
Currently, Super Agent has launched public testing. From models to protocols, tool ecosystems, and personalized task orchestration, Nano AI seems to be aiming for a product-level innovation that truly brings AI agents into everyone’s daily life.
So, how good is Nano AI’s ‘MCP Toolbox’? To answer this question, the Machine Heart team, having gained internal testing qualifications, conducted a series of tests.
Hands-on Experience with the Toolbox: MCP Made Simple
Using the Nano AI ‘MCP Toolbox’ has a very low barrier to entry. Users only need to download and install the Nano AI application and then register and log in, without any additional configuration.
Upon entering the updated ‘Agent’ page, we can see that Nano AI has categorized existing agents into several broad categories, including in-depth research, work and efficiency, and life assistants. At the same time, it also provides access to the Toolbox and a case study square.
Entering the Toolbox, we can see that Nano AI has already configured over 100 MCP Servers (this number increased from 120 to 132 during the writing of this article), including a dozen MCP tools developed by Nano AI itself and hundreds of third-party MCP tools, covering various scenarios such as office collaboration, academics, life services, search engines, finance, media entertainment, and data crawling, making it the largest MCP ecosystem in China. Additionally, Nano AI also supports users in configuring their own MCP Servers. In the following, we will use the term ‘Tool’ instead of ‘MCP Server,’ and the reason for this will be explained in detail later.
First, let’s test an application that Machine Heart readers will find most appealing: searching and organizing recent research findings on arXiv related to a specific research topic.
Let’s first search the Toolbox and find that Nano AI’s preset tools already include ‘arXiv Search,’ so we don’t need to configure it ourselves. Looking back, we can also see that Nano AI already has many agents that support arXiv paper retrieval. We will choose ‘Professional Paper Search’ as our first step. We can see that this agent is configured with four tools: Nano AI Super Search, arXiv Search, Google Scholar, and Academic Search, which perfectly meets our needs. Write a prompt and execute:
Retrieve research findings related to reinforcement learning on arXiv in the past month, classify them according to theoretical research, technological improvements, and applications, and provide a simple interpretation of the important progress.
The working process of ‘Professional Paper Search’ is as follows:
This task is very simple. The agent only called the ‘arXiv Search’ tool once, and therefore completed the task in less than half a minute, selecting two representative research results in each of the three categories.
Next, try the cycling planner agent using the command: “Are there any good cycling routes near Guanyin Bridge in Chongqing?”
We can see that this agent used three tools: amapmcpserver-cloud’s maps_weather (for querying weather) and maps_direction_bicycling (for setting routes) and gen_html (for generating webpages), executing for a total of 362 seconds, and finally obtained the dynamic webpage shown above. You can also access it through this link: . Yes, you can publicly share the generated webpage!
Next, let’s increase the difficulty. This time our requirement is “Search the network, analyze the current women’s fashion trends, and issue a women’s fashion element analysis report.” This time we will directly use Nano AI’s ‘In-Depth Research Agent,’ which can choose to use appropriate tools according to the user’s specific needs, including MCP Servers and the built-in browser to complete various computer-use tasks. Of course, therefore, the In-Depth Research Agent often takes much longer to execute a task, up to tens of minutes.
When executing the task, the In-Depth Research Agent will first plan the steps to be executed according to the task requirements, and then execute the steps step by step according to the plan.
The execution steps generated by the In-Depth Research Agent for this specific task are shown in the figure below.
First, it searched for content related to the current women’s fashion trends on multiple websites, then analyzed the searched content, and visualized the results. Finally, it gave the final report.
In this process, it called the local search tool aiso_do_search three times, the data crawling tool 360_crawl once, the cloud code sandbox tool cloud-sandbox nine times, the summary tool summary once, and the webpage generation tool gen_html once.
In the end, we obtained a 30-page in-depth report, covering six major sections: popular style theme analysis, popular color trends, popular styles and element analysis, comprehensive evaluation of popular elements, fabric and technology trends, and matching suggestions and applications, far exceeding our initial one-sentence task.
Several pages of content extracted from the report
The following video shows the entire process of Nano AI’s In-Depth Research Agent completing the task:
Played at 4x speed
Not only that, Nano AI also generated a dynamic webpage that can more vividly display the analysis results obtained:
In addition, considering that Google recently released its first-quarter financial report, we can also let Nano AI’s ‘Chief Industry Insight Officer’ agent help us interpret it.
Its webpage version can be accessed at: , and the entire working process can be seen in the following video:
Let’s try using Nano AI to write a movie review suitable for posting on Xiaohongshu for the recently popular TV series ‘The Good Life’, and the preset Xiaohongshu browsing robot can do the job well.
Beware! The content will contain spoilers.
The following video shows the entire process of Nano AI working.
Wecan see that in this process, Nano AI used two tools related to Xiaohongshu, including collect_relate_info_redbook for collecting information on Xiaohongshu and red_book_generate for generating Xiaohongshu content; in addition, it also used browser_automation_task - this tool can open the built-in browser in the Nano AI application to perform tasks. With the appropriate instructions, you can also use this tool to complete tasks such as booking train tickets, posting on Weibo, and taking notes in one sentence.
Finally, on Nano AI, users can also easily configure their own MCP. For example, here, we successfully configured a tool for querying and analyzing Obsidian notes with just a few parameter settings.
Then, just configure an agent that calls the tool, and we can intelligently retrieve and analyze our collected notes in Nano AI. The following video shows an example:
The above cases are just the tip of the iceberg of Nano AI’s capabilities. With the MCP Toolbox, there are many other things users can do, such as crawling and searching information, generating images and video content, letting AI organize your flomo fragment notes and put the results into the Notion workspace, analyzing stocks, finding the most cost-effective flight route to travel to Portugal, specifying travel or fitness plans, creating company reports, managing cloud storage repositories or local files… The only limit is your imagination!
Hiding MCP in the Toolbox: How Nano AI Does It
MCP, or Model Context Protocol, was first released by Anthropic in November 2024. It can be said to be an important ‘bridge’ connecting large models with the real world - it allows models to not only answer questions, but also call tools, obtain data, and execute tasks like humans. This year, as more and more companies adopt the protocol, it has become a de facto standard in LLM’s use of tools, further releasing the potential of AI agents.
However, for most users, the typical labels of the MCP protocol are ‘complexity’, ‘high technical threshold’ and ‘developer exclusive’. How to hand over this ability, which originally belonged to professional engineers, to every ordinary person?
In response to this real problem, 360’s answer is: no longer teach you to understand MCP, but directly encapsulate it into a set of ‘visible, clickable, and result-predictable’ toolbox.
1. From Concept Simplification to Interaction Dimensional Reduction
The Nano AI team first did the translation of concepts: users do not need to understand what MCP Server or API Key is, they only need to know that this is a usable ‘tool’ or ‘skill’ - which is why we used the term ‘tool’ earlier. Packaging the originally obscure protocol interface into easy-to-understand tool labels such as ‘search’, ‘writing’, and ‘data analysis’ greatly reduces the user’s cognitive threshold and allows users to more intuitively understand the meaning of the so-called MCP Server to AI large models. This is the design philosophy of the Nano AI Toolbox. Behind this is Nano AI’s re-encapsulation of the MCP protocol and the engineering reconstruction of the interface layer.
What users see in the interface is simple selection and dragging, but in reality, it is scheduling more than 100 MCP Servers developed by Nano AI itself or a carefully selected integration. These tools cover scenarios such as office, academics, finance, search engines, web crawling, and image processing. Users can let large models automatically call these ‘external brains’ to complete complex task chains without writing a line of code.
Nano AI even has built-in API Keys for multiple MCP tools such as Firecrawl, Brava Search, and AutoNavi Maps.
2. Breaking Through the ‘Last Mile’ Between Models and Tools
In the past, even if large models had powerful language understanding capabilities, they were still trapped in the ‘tool calling’ island effect. Nano AI’s approach is to use the MCP protocol as an intermediary language, fundamentally breaking through the collaboration mechanism of ‘large model + tool’.
This not only solves the problem of calling but also greatly expands the actual ability boundary of the model. For example, users only need to tell the agent “Help me generate a NVIDIA stock price analysis report,” and the agent can automatically break down the task steps, mobilize search engines, crawl page content, generate analysis charts, and output a clearly structured report. During the period, 5 to 7 tools may be called, but the user only sees one result page.
This is precisely the embodiment of MCP’s ‘tool combination’ ability: it allows agents to independently schedule resources, plan processes, and conduct trial-and-error feedback and self-optimization during operation, forming a highly anthropomorphic task-solving path.
3. Local Operation, Safe and Reliable: In-Depth Polishing of Technology Stack
Unlike many ‘cloud intelligent bodies’, Nano AI chose a more difficult but more promising path: deploying MCP clients locally, giving users greater control.
This brings at least three key advantages:
- Call freedom: Local intelligent bodies can access the user’s file system, call the browser, and retrieve the database to achieve true personalized task processing.
- Breaking through barriers: In response to the unique needs of AI, 360 has created a dedicated AI browser for Nano AI and adapted it to the mainstream platforms in China. It can break through login walls, man-machine verification, and information flow interference, and automatically complete operations such as login and sliding verification.
- Sandbox guarantee: Based on 360’s security technology accumulation, Nano AI will also introduce a local runtime sandbox in the future, which can monitor, early warn, and restrict the large model from possibly misoperating local files in real-time to ensure data security.
This whole system not only allows users to ‘use’ it, but also ‘use it safely, efficiently, and scalably.’
4. Facing Massive Users: Building a Truly Open MCP Ecosystem
Nano AI not only encapsulates MCP tools but also took the lead in opening up an open skill ecosystem. At present, this platform with a monthly visit volume of more than 400 million has more than 100 high-quality MCP tools online, and more third-party MCP Servers are being entered. Users can freely upload, reuse, and combine tool skills to create their own AI agent.
For ordinary users, this means that it is no longer ‘using AI set by others’ but can build a personalized AI assistant according to their own needs. Paper analysis, data generation, trend monitoring, webpage construction, stock prediction… As long as there is a demand, there are tools that can be used in combination, and there are tasks that can be executed automatically.
For the entire industry, this means that agent technology is moving from the ‘closed system’ to the ‘ecological network’ stage. Tools, models, and tasks will no longer be isolated but will be linked by MCP as a common language, creating an unprecedented intelligent collaboration pattern.
Technical Barriers Have Been Broken: Intelligent Bodies Sink to the C End
Once upon a time, the threshold for using intelligent bodies was still high on the door frame of developers. Now, with the launch of Nano AI ‘MCP Toolbox’, MCP, a protocol known as AI automation infrastructure, has entered the vision of ordinary users for the first time in an almost ‘fool-style’ form. As Zhou Hongyi, chairman of 360 Group, said at the sharing meeting before the release: ‘What MCP Server is automatically called in the agent, users don’t need to know.’ With the toolbox, Nano AI is breaking the technical barriers of MCP and allowing intelligent bodies to further sink to the C end.
Making MCP into a ‘toolbox’ sounds easy, but it is difficult to do. This not only tests the ability to integrate technology, but also tests the ‘empathy’ of product thinking and user understanding. What Nano AI is doing is to encapsulate complexity in the core and give freedom to users - so that every ordinary person can have the permission to ‘call the AI world’ like developers.
This process is not a simple visual interface construction, but a deep AI application paradigm change: intelligent bodies are no longer just models that can speak and answer, but real partners with the ability to schedule capabilities, call tools, and complete tasks.
Since then, MCP has truly begun to move towards C-end users, which may be a historical starting point worth remembering.