unfetch
A download manager for humans and AI with a native MCP server: add_task, list_tasks, wait_for_task with progress notifications from any MCP host. Multi-threaded multi-mirror HTTP, BT/magnet, 1000+ video sites (yt-dlp), RSS, token-secured remote Web UI. ~8MB, open source.
README
<div align="center">
unfetch
A modern download manager for humans and AI
Website · Download · Sponsor · 中文
</div>
unfetch is the download engine that finishes what your browser starts. Fast multi-mirror HTTP, real BT/magnet support, 1000+ video sites via yt-dlp, RSS subscriptions, completion hooks, and a token-secured remote Web UI — all driven by GUI, CLI, or AI through MCP. ~8 MB installer. MIT-style license. Zero telemetry.
Features
- Multi-mirror HTTP — up to 32 threads × multiple mirror URLs, automatic fallback, resume, SHA256/MD5 verification.
- BT / Magnet — 35 public trackers, IPv6 + WebTorrent, file picker before download, peer/seeder stats, auto uTP fallback when your ISP throttles.
- 1000+ video sites — YouTube, Bilibili, TikTok, X, and the rest via bundled yt-dlp.
- AI-native (MCP) —
add_task,list_tasks,wait_for_task, progress notifications — from any MCP host (Claude Desktop, Cursor, Codex, …). No vendor lock-in. - Remote Web UI — token-secured Web UI on
0.0.0.0— manage downloads from your phone or remote box. QR-shareable. - RSS auto-fetch — subscribe with regex filters, new items queued automatically, GUID deduped.
- Completion hooks — webhook POST or shell exec on done — pipe into Home Assistant / n8n / anything.
- Task templates & dependency chains — per-site Cookie/UA presets, wait-for-other-task, scheduled start times, per-hour bandwidth schedule.
- 6 themes — Cream Light (default for new users), Deep Indigo, Light Paper, Forest Night, Cyberpunk, Mono Gray — switch live without restart.
- Privacy first — no ads, no telemetry, no login, no infohash uploads.
Install
| Platform | Download |
|---|---|
| macOS (Apple Silicon + Intel, Notarized) | unfetch_*_universal.dmg |
| Windows 10/11 (NSIS installer) | unfetch_*_x64-setup.exe |
| Linux x64 (AppImage / deb / rpm) | Releases page |
Or grab the latest from https://unfetch.org/#download.
Three frontends, one daemon
unfetch ships as a Go daemon + a Tauri shell. Three frontends share one daemon, all backed by the same REST API on localhost:19543:
- GUI — Tauri window (the default app icon launches this)
- CLI —
unfetch add <url>,unfetch list,unfetch pause <id>, … - MCP —
unfetch --mcp(stdio JSON-RPC, drop into Claude Desktop config) - Browser extension — see
extension/for Chrome/Edge sniffer
┌──────────────────────────────────────────────────────────┐
│ Tauri 2 app (Rust + React + 6 themes) │
│ ├─ Main window / mini floating window │
│ ├─ Tray, clipboard sniffer, deep-link handler │
│ └─ Forwards arguments / events to the daemon │
└─────────────────────────────┬────────────────────────────┘
│ localhost:19543 (REST + SSE)
▼
┌──────────────────────────────────────────────────────────┐
│ Go daemon (anacrolix/torrent + http downloader + yt-dlp) │
│ ├─ SQLite store of all tasks (resumes across restarts) │
│ ├─ Speed scheduler, retry queue, hash verifier │
│ ├─ Remote Web UI on 0.0.0.0 (token-gated) │
│ └─ Same API consumed by GUI / CLI / MCP / extension │
└──────────────────────────────────────────────────────────┘
Use with AI (MCP)
Drop this into your Claude Desktop / Cursor config (~/Library/Application Support/Claude/claude_desktop_config.json on macOS):
{
"mcpServers": {
"unfetch": {
"command": "/Applications/unfetch.app/Contents/MacOS/unfetch",
"args": ["--mcp"]
}
}
}
Then your AI can add_task, list_tasks, wait_for_task, etc. See docs/mcp.md for the full tool schema, or visit https://unfetch.org/mcp.
Build from source
Prerequisites: pnpm, Rust toolchain, Go 1.23+, Node 20+.
# 1. Frontend
pnpm install
pnpm build
# 2. Go daemon (per-platform — Tauri externalBin requires platform-suffixed names)
cd core
GOOS=darwin GOARCH=arm64 go build -ldflags="-s -w" \
-o ../src-tauri/binaries/unfetch-daemon-aarch64-apple-darwin .
# (repeat for x86_64, linux, windows targets)
# 3. Tauri release build
cd ..
pnpm tauri build # native target only
pnpm tauri build --target universal-apple-darwin # macOS universal
For macOS release builds with notarization, see scripts/build-mac.sh.
Release workflow
./scripts/release.sh 0.3.0 # bump → commit → tag → push (triggers CI for Win/Linux)
./scripts/build-mac.sh 0.3.0 # local universal build + Apple notarize + upload dmg
CI in .github/workflows/release.yml only builds Windows + Linux — macOS is notarized locally because Apple Developer credentials should not live in CI.
Support development
unfetch is built and maintained by one developer. If it saves you time, sponsorship helps keep it alive:
- GitHub Sponsors — international, monthly or one-time (Stripe)
- PayPal — worldwide one-time tip
- Alipay / WeChat Pay — scan QR at https://unfetch.org/#sponsor
License
MIT-style license. See LICENSE (forthcoming — currently de-facto MIT).
Contributions to upstream are encouraged — unfetch's maintainer is also an active contributor to anacrolix/torrent.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。