Fans publish community mod that restores Memory, Skills and MCP to DeepSeek V4 web client

Community mod fills missing features

A group of Chinese users has reportedly released a community modification of the web version of DeepSeek (深寻) V4 that restores or implements several popular features long requested by users — notably Memory, Skills and an MCP-like control panel. The mod appears aimed at filling functional gaps left by the official V4 web release, offering an experience closer to what advanced Agent workflows require. It has been reported that the patch is circulating on Chinese developer forums and Git repositories; the changes have not been validated by DeepSeek itself.

Why users moved to modding

Why did volunteer developers step in? Short answer: demand. Agent-style workflows consume far more tokens and rely on persistent state (Memory), modular task tools (Skills) and management/control interfaces (MCP). As commercial model providers race to lower inference costs and scale Agent calls, many users find the official web UI lags behind in practical tooling. The fan mod suggests an appetite for hands-on features that enable longer-running, multi-step Agents — and an impatience with waiting for vendor roadmaps.

Commercial and geopolitical backdrop

The mod arrives as DeepSeek’s broader strategy appears to be shifting. It has been reported that Contemporary Amperex Technology Co., Limited (CATL) (宁德时代) has decided to join DeepSeek’s first funding round, with other potential investors reportedly including Tencent (腾讯), JD.com (京东), NetEase (网易), IDG and Monolith. The company also announced a permanent 75% cut to V4-Pro API pricing on May 23. Behind the scenes sits Huanfang Quant (幻方量化), the quantitative firm that incubated DeepSeek; the firm’s large capital base and focus on engineering-heavy deployment help explain the platform-style shift. Domestic hardware adaptation (国产卡适配) and efforts to reduce inference costs must also be read in the context of broader US-China technology decoupling and supply-chain pressures.

Implications and unanswered questions

The community mod is a reminder that user ecosystems can move faster than corporate product cycles. Will DeepSeek accept community contributions, or will the company double down on centralized platform control as it chases scale and institutional investors? There are risks too — IP, security and regulatory scrutiny increase when third parties modify deployed AI front-ends. For now, the mod underscores a central tension in the Chinese AI market: model capability matters, but so does the tooling that unlocks real-world Agent usage — and the race to capture that usage is increasingly an infrastructure play.