MCP SuperAssistant browser plugin: let ChatGPT, Gemini, DeepSeek web directly call MCP tools (mit open source)MCP SuperAssistant 浏览器插件：让 ChatGPT、Gemini、DeepSeek 网页版直接调用 MCP 工具（MIT 开源）

The biggest weakness of the web version of AI is "not enough": the model is no longer smart, and it can't read your local code repository, check your database, and tune your company's internal interfaces.

MCP (Model Context Protocol) was originally an open protocol to solve this problem, but the client mainly falls on the desktop and the IDE, and there has been a lack of a common bridge layer in the browser.

This extension makes the bridge a “visible button”: when the AI writes a tool call in the answer, the page renders it as a clickable execution block, without having to manually copy the parameter and paste it on the command line.

Typical scenarios: let ChatGPT web directly read your PR on GitHub, let Gemini check the local SQLite, and let DeepSeek call your own knowledge base retrieval, all in a conversation stream.

ChatGPT, Google Gemini, Google AI Studio, Perplexity, Grok, DeepSeek, Kimi, Qwen Chat, Z Chat, GitHub Copilot, Mistral AI, OpenRouter Chat, T3 Chat, and more.

It is injected into the page through content scripts and does not rely on various official plug-in interfaces, so the usability is less affected when the platform is revised or the API policy is changed.

Chrome, Edge and other Chromium core browsers can be installed; the official Chrome Store and Firefox Add-ons are also available, or you can build your own.

Automatic tool recognition: Scan the MCP tool call format in the AI response and highlight it directly on the page.

One-click execution: Click the execution block to call the native MCP service, no need to switch windows or manually spell commands.

Result backfill: The execution result is automatically inserted into the dialog box, and the AI can continue to reason down based on the real data.

Multi-platform parallelism: ChatGPT and Gemini are turned on at the same time, and different MCP server combinations can be mounted on both sides.

The following is the complete source code package of the project's official repository. The chrome-extension directory in the compressed package is the extension ontology. The repository is managed by pnpm workspace, and you need to build a loadable product directory yourself.

The most convenient way is to search for MCP SuperAssistant one-click installation directly in the Chrome App Store; if you want to change the code, or use a submission that is newer than the store, take the source code route.

Source code building process: Install Node.js 20 and above with pnpm, execute pnpm install in the repository root directory, then enter the chrome-extension directory to execute the build command, and the product will be output to the mv3 directory under .output.

Open chrome://extensions, enable "Developer Mode" in the upper right corner, click "Load Extracted Extensions" and select the output directory in the previous step.

After installing the point extension icon, fill in the MCP service address (native HTTP or SSE endpoint), refresh the AI web page, and the tool execution panel will appear on the page.

The extension itself does not upload data, but the conversations on the AI web will still be sent to the large model service provider of your choice. If the page involves sensitive information such as accounts and keys, please be careful to let the AI read it. Executing the MCP tool is equivalent to letting the model operate your native resources indirectly. It is recommended to first test the water with a read-only tool, confirm the behavior, and then release the write permission. This article is compiled from self-disclosed materials, and the resources are for learning and communication only. Please follow the project mit permission.