zhu1090093659/deepseek-pp
DeepSeek Web browser extension: AI agent workspace with MCP tools, memory, Skills, automation, web search, and conversation export.
1,470 ★186 forksTypeScriptUpdated 22h ago
What you need to know
Open-source browser extension (Chrome/Edge/Firefox) that turns DeepSeek's web app into a bilingual (Simplified Chinese / English) AI agent workspace with long-term memory, projects, Skills, MCP tools, multimodal media analysis, browser control, conversation export, and scheduled automation.
Install
Install from Chrome Web Store (chrome web store listing 'DeepSeek++'), then open chat.deepseek.com git clone; npm install; npm run build (default Chrome MV3); npm run build:chrome|edge|firefox|all; load unpacked from dist/ npx deepseek-pp-multimodal-mcp install --browser chrome --extension-id <ID> # multimodal Native Host npx deepseek-pp-shell-host install --browser chrome --extension-id <ID> # Shell Native Host + command-line OfficeCLI
Usage
- •Use the sidebar to enable memory, Skills, MCP, web tools, conversation export, and automation; sidebar chat supports DeepSeek login or official API key
- •Sidebar MCP page: add remote/local MCP services, authorize, test connections, refresh tools; MCP tool results feed back into the same session for agent-style multi-step execution
- •Multimodal: after installing the Native Host, attach images (via OpenAI) or videos (via Gemini) in the DeepSeek input box; configure keys in Settings -> Multimodal API
- •OfficeCLI: create a Shell preset in the sidebar so the model calls officecli via shell_exec for checking/validating/modifying .docx/.xlsx/.pptx
- •Type / in the chat box for Skill autocomplete; create automation tasks with immediate run, cron, or RRULE schedules (min interval 15 min)
Key features
- ✓MCP tool system with remote/local servers, per-service and per-tool permission management, and result return to the session for continued generation
- ✓Long-term memory (user/feedback/topic/reference types) with keyword-based relevance injection; Skills support /-triggered autocomplete, custom, GitHub import, and local folder import + sync
- ✓Multimodal media analysis (OpenAI for images, Gemini for video) via a Native Host; browser control on user-selected tabs (navigation, click, input, file upload) in Chrome/Edge
- ✓Project context, downloadable single-file and project-bundle artifacts, and conversation export (HTML/Markdown/PDF) with attachment manifests
- ✓Automation tasks in independent DeepSeek sessions with cron/RRULE scheduling, status tracking, and agent-style continuation
- ✓Bilingual UI (follow browser / Simplified Chinese / English) that keeps user-created memories, presets, and skills untranslated; desktop pet ('DeepSeek whale')
Best for
DeepSeek web users who want to keep chatting in the official web app while adding MCP tools, memory, skills, multimodal analysis, browser automation, and scheduled tasks.
Caveats
- ⚠PC browsers only: officially supports Chrome, Edge, and Firefox — no Android or other mobile builds
- ⚠Browser control is only available in Chrome/Edge; platform boundaries are documented in the UI
- ⚠MCP configs and keys are stored in browser local storage; cloud sync does not sync sensitive information
- ⚠Unofficial third-party extension overlaying DeepSeek's web app (relies on the web session or an official API key); Apache-2.0
Platforms: Web (Chrome) · Web (Edge) · Web (Firefox)
Reviewed 2026-08-07
Topics
agentic-aiai-agentai-memoryautomationbrowser-extensionchrome-extensionconversation-exportdeepseekdeepseek-aiedge-extensionfirefox-extensionmcpmodel-context-protocolnative-messagingofficeclireacttool-callingtypescriptweb-searchwxt
- Stars
- 1,470★
- Forks
- 186
- Language
- TypeScript
- License
- Apache-2.0
- Created
- 2026-04-27
- Last push
- 2026-08-06