Install URL is here：https://github.com/lobehub/lobe-chat/releases

---

After 2 to 3 months of continuous development, we have finally completed the biggest project in LobeChat's history: MCP Marketplace & one-click desktop installation! Without further ado, let's quickly check out a full demonstration:

Blender.MCP.mp4

Feature Upgrades

Comprehensive Upgrade of the Plugin Store

In terms of design, we drew inspiration from popular plugin marketplaces (such as VSCode, Webstorm, etc.) and conducted a comprehensive upgrade of the plugin marketplace to reveal more abundant and richer information

Note: Currently, the MCP plugin tab is only available on the desktop version

One-click installation of MCP plugins

Based on the updates in the plugin marketplace, we now support one-click installation for all Local type MCP plugins—just click to install:

MCP.One.Click.Install.mp4

Complete Revamp of the Discovery Page

In Discovery, we have fully upgraded the style for MCP plugins, revealing more information and supporting more types of sorting.

Data Collection Feedback Request

Starting from this feature release, our subsequent development features will gradually incorporate data collection to better analyze user behavior and help us further iterate and optimize the product.

For the MCP-related features, we will record two types of data regarding the use of MCP plugins: installation and invocation.

For plugin installation, we will collect:

• Plugin installation id, version, manifest
• Installation duration
• Installation command (command and params, excluding env)
• Whether the installation was successful
• If failed, the error code and reason for failure, and record the complete related error log in metadata;
• Installed device unique id, device platform type, request IP, and userAgent

For plugin calls, we will collect:

• The plugin id being called,
• The tool name of the plugin being called, the input and output of the call
• Whether the call was successful
• If it fails, the error code and reason for the failure
• Session ID
• Unique device installation ID, device platform type, request IP, and userAgent

By recording the above data, we can compile statistics such as the number of MCP plugin installations, installation success rate, number of calls, and call success rate, serving as part of the evaluation system for the MCP plugin. This will help users select high-quality MCP Servers and also assist developers in better understanding how their MCP Servers are being used, enabling iteration and optimization.

However, data collection inevitably involves sensitive factors, so we want to be open and transparent here about the scope of our data collection and seek your opinions and feedback. For example, which types of data collection are you more concerned about, which you think should not be collected, or which data should be collected to promote ecosystem development. We are very eager to hear your thoughts!

Note: The current data collection logic does not have a toggle switch yet and is enabled by default. I will add this feature in iterations before the official release

---

经过累计长达 2~3 个月时间的开发，我们终于完成了 LobeChat 有史以来最大的一个项目： MCP Marketplace & 桌面端一键安装！ 话不多说，快速来看一个完整演示：

Blender.MCP.mp4

功能升级

插件市场全面升级

在设计上我们借鉴了市面上流行的插件市场（例如 VSCode、Webstorm 等），针对插件市场做了一次全面升级，以透出更多更丰富的信息

注：目前 MCP 插件 tab 只在桌面端可用

一键安装 MCP 插件

在插件市场的更新基础上，我们支持了所有 Local 类型 MCP 插件的一键安装，只需点击安装即可：

MCP.One.Click.Install.mp4

发现页全面焕新

在发现，我们针对 MCP 插件做了风格的全面升级，透出了更多信息，也支持了更多类型的排序。

使用示例：playwright MCP 插件自动发推

我装了 playwright MCP 插件后，就直接可以实现自动发推的效果：

Twitter.mp4

其中填写账号密码都是我自己操作的，然后剩下的都交给 AI 即可（而且哪怕是 gpt-4.1-mini 这样的小模型，现在的 agentic 能力都足以做到！）。

twitter 链接：https://x.com/arvin17x/status/1942481767128260736

数据采集意见征询

从本次功能上新开始，我们的后续开发功能都会逐步加入数据采集，以便我们更好地分析用户的行为，帮助我们进一步迭代优化产品。

在本次 MCP 相关的功能上，我们针对 MCP 插件的使用会记录两类数据：安装和调用。

针对安装插件，我们会采集：

• 安装插件 id、版本、manifest
• 安装耗时
• 安装指令（cmmand 和 params，不含 env）
• 是否成功安装
• 如果失败，失败的错误码和错误原因，以及在 metadata 中记录相关的完整错误日志；
• 安装设备唯一 id、设备平台类型与 userAgent

针对调用插件，我们会采集：

• 调用插件 id，
• 调用插件的 tool 名
• 是否调用成功
• 如果失败，失败的错误码和错误原因
• 会话 id
• 安装设备唯一 id、设备平台类型、userAgent

通过记录以上数据，我们将可以统计出来 MCP 插件的安装次数、安装成功率、调用次数、调用成功率等信息，作为 MCP 插件的评价体系一环。如此一来就可以帮助大家筛选优质 MCP Server，同时也能进一步帮助开发者更好地了解自己做出来的 MCP Server 是怎么被使用的，并进行迭代与优化。

但数据采集总归是有敏感因素在，所以也在这里和大家开诚布公地说明下我们的采集范围，并且征求下大家的意见和反馈，比如大家对于哪些数据的采集会比较在意，觉得不应该采集，或者说哪些数据是应该被收集以促进生态发展的，非常希望听听大家的看法！