In the highly competitive field of large-scale language modeling, when the market's eyes are focused on Musk's Grok
and the Dark Side of the Moon Kimi
time, Alibaba quietly launched a desktop client for its Qwen family of models, further expanding its AI ecosystem.
together with OpenAI
cap (a poem) ChatGPT
The relationship is similar in that Ali and Tongyi Qianqian are the R&D entities and product brands behind them. However, Ali's strategy shows clear internal and external differences. Its domestic version, Tongyi Qianwen (tongyi.com), is a feature-rich platform that integrates a variety of AI Agent functions such as document processing, presentation creation, and voice transcription, but with less freedom in model selection.
In comparison, its overseas version Qwen Chat
(qwen.ai) has adopted a more proximity-based approach to ChatGPT
maybe Claude
s clean design style. It retains core features such as deep research, networked search and image generation, and allows users to Qwen
The series can be freely switched between different models, and it is even possible to have multiple models answer the same question for comparison.
This release of the desktop client is a great way to Qwen Chat
Designed to provide macOS
Version.
download address :: https://qwen.ai/download
Once installed and logged in, the client interface is highly similar to the web version, but with the addition of a key option below the input box: MCP
The
MCP
The model context protocol (Model Context Protocol), which was created by Anthropic
An open source standard proposed to allow AI models to interact more flexibly with external tools and data sources. In a nutshell. MCP
Like a "universal socket", it allows the Qwen
Such a language model connects in real time to external applications or databases and no longer relies solely on its inherent training data. This mechanism is similar to the OpenAI
(used form a nominal expression) GPTs
maybe Actions
The goal is to transform the big model from a closed "brain" into an open platform where tools can be invoked to solve real-world problems.
exist Qwen Chat
In the client, you can manage the official built-in plug-ins through "Set MCP", including:
- code-interpreter : Used to execute and interpret code.
- fire-crawl : Provide networked search and real-time information analysis capabilities.
- amap : Integration of Gaode map service for searching geographic information and planning routes.
- image-generation : Call AI for image generation.
Users can enable or disable these plugins with a single click. Once enabled, the plug-ins will be disabled when working with the Qwen Chat
During the conversation, the model determines whether these external tools need to be called to optimize the answer.
For example, when asking "What is the distance from Shanghai to Beijing? MCP
(used form a nominal expression) Qwen Chat
It will automatically call the Goldmap plugin in the maps_distance
et al. (and other authors) API
, providing precise answers based on real-time geographic data, not on vague memories from training data.
In addition to the officially provided plugins, users can also add customized MCP
Service. There are three sample localization plugins built into the client:
- Fetch : Crawls the content of a given web page and converts it into a
Markdown
Format. - Filesystem : Authorized models access the local file system to perform operations such as reading and writing files.
- Sequential-Thinking : Bootstrapping models use a structured, step-by-step approach to complex problems.
With the "Add MCP" button in the upper right corner, advanced users can connect to their own developed MCPs. MCP
server, thus extending the functionality of the model indefinitely.
It is worth noting that Qwen Chat
It also provides a function called Canvas
The function. It is a whiteboard or canvas that allows users to freely combine and iterate the outputs of multiple models on it, visually organizing different ideas and results, which is appealing to users who need to brainstorm or create complex content.
Qwen Chat
The desktop side of the rollout, and in particular, its impact on the MCP
The integration shows Alibaba's strategic thinking at the AI application level. Not only is it completely free, offering a rich selection of models and powerful scalability, but more importantly, it further hands the control and customization of models back to the user. Although it currently only supports macOS
platform, but this attempt to build an ecosystem through open protocols gives it a unique advantage and a high degree of playability in the fierce competition for AI products.