AI Skill Hub 推荐使用:优先级锻造 是一款优质的MCP工具。AI 综合评分 7.2 分,在同类工具中表现稳健。如果你正在寻找可靠的MCP工具解决方案,这是一个值得深入了解的选择。
优先级锻造 是一款遵循 MCP(Model Context Protocol)标准协议的 AI 工具扩展。通过 MCP 协议,它可以让 Claude、Cursor 等主流 AI 客户端直接访问和操作外部工具、数据源和服务,实现 AI 能力的无缝扩展。无论是文件操作、数据库查询还是 API 调用,都可以通过自然语言在 AI 对话中直接触发,极大提升生产效率。
优先级锻造 是一款遵循 MCP(Model Context Protocol)标准协议的 AI 工具扩展。通过 MCP 协议,它可以让 Claude、Cursor 等主流 AI 客户端直接访问和操作外部工具、数据源和服务,实现 AI 能力的无缝扩展。无论是文件操作、数据库查询还是 API 调用,都可以通过自然语言在 AI 对话中直接触发,极大提升生产效率。
# 方式一:通过 Claude Code CLI 一键安装
claude skill install https://github.com/Unobtainiumrock/priority-forge
# 方式二:手动配置 claude_desktop_config.json
{
"mcpServers": {
"-----": {
"command": "npx",
"args": ["-y", "priority-forge"]
}
}
}
# 配置文件位置
# macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
# Windows: %APPDATA%/Claude/claude_desktop_config.json
# 安装后在 Claude 对话中直接使用 # 示例: 用户: 请帮我用 优先级锻造 执行以下任务... Claude: [自动调用 优先级锻造 MCP 工具处理请求] # 查看可用工具列表 # 在 Claude 中输入:"列出所有可用的 MCP 工具"
// claude_desktop_config.json 配置示例
{
"mcpServers": {
"_____": {
"command": "npx",
"args": ["-y", "priority-forge"],
"env": {
// "API_KEY": "your-api-key-here"
}
}
}
}
// 保存后重启 Claude Desktop 生效
AI-powered task prioritization that learns from your decisions. Cross-project tracking with heap-based priority scoring, drag-and-drop reordering, and online ML that adapts to how you work.

Real-time priority queue with expandable task details, workspace switching, and heuristic weight tuning
</div>
<details> <summary><strong>📸 More Screenshots</strong></summary>
<br>
#### Priority Queue with Expandable Tasks 
Click any task to reveal notes, dependencies, and completion controls
#### Workspace Switcher 
Separate task databases for work, personal, client projects—ML learning persists across all
#### Heuristic Weight Tuner 
Real-time sliders to adjust how blocking dependencies, cross-project impact, and time sensitivity affect priority scores
</details>
---
npm install
1. Install WSL2 + Ubuntu: in PowerShell as Administrator —
wsl --install -d Ubuntu
Reboot when prompted. Open Ubuntu from the Start menu and create your Linux user. 2. Enable systemd in WSL2 (required for the persistent backend/frontend services). Inside Ubuntu: sudo tee -a /etc/wsl.conf >/dev/null <<'EOF'
[boot]
systemd=true
EOF
Then, in PowerShell: wsl --shutdown
Reopen Ubuntu. Confirm: systemctl --user status should not say "Failed to connect to bus". 3. Install Node ≥20 inside WSL (the Windows-side Node install does NOT count — the stack runs as a Linux process). nvm is the easy path: curl -o- https://raw.githubusercontent.com/nvm-sh/nvm/v0.40.0/install.sh | bash
exec bash
nvm install 20
```bash
bash setup.sh ```
The setup script will: 1. Check for and install prerequisites (Node.js via Homebrew or nvm) 2. Install npm dependencies 3. Initialize your task database 4. Interactively configure your AI tool (Cursor, Droid, Claude Code, or Codex) 5. Verify everything works
If you prefer manual control:
```bash
```bash
./setup.sh install-launchd ```
This will: - Copy launch agent files to ~/Library/LaunchAgents/ - Update paths to match your installation - Load and start all three services (backend, frontend, watchdog) - Verify they're running correctly
./setup.sh uninstall-launchd ```
The deployed plists contain the absolute path to the node and npm binaries that were on PATH when the installer ran. If you migrate between Homebrew / the official .pkg installer / nvm / Volta / fnm / asdf, re-run ./setup.sh install-launchd so the plists are regenerated with the new binary paths.
If you prefer manual control, see detailed instructions in launchd/README.md.
Note: Launch agents only work on macOS. For Linux, see below. For Windows, see Windows Support.
---
./setup.sh install-systemd
This copies unit files to ~/.config/systemd/user/, enables linger so services start at boot (not just at login), and starts all three services immediately.
bash setup.sh # Full setup (prerequisites, deps, MCP config) npm run setup:mcp # Configure MCP for your AI tool (interactive) npm run seed # Initialize/reset database npm run verify # Verify setup is working
Critical: clone into your WSL home, NOT /mnt/c/....
cd ~
git clone https://github.com/Unobtainiumrock/priority-forge.git
cd priority-forge
./setup.sh install
Why not /mnt/c/? WSL's /mnt/<drive>/ mounts use DrvFs, which is roughly 100x slower than ext4 for SQLite writes and is unreliable for the chokidar file-watcher used by Vite and tsx watch. The stack will appear to install, then silently miss reloads and corrupt data on concurrent writes. setup.sh and npm run setup:mcp will refuse to run from /mnt/<drive>/ paths; override with PRIORITY_FORGE_ALLOW_CROSS_MOUNT=1 only for read-only smoke tests.
For team use with shared tasks, deploy to a shared server:
{
"mcpServers": {
"priority-forge": {
"type": "http",
"url": "https://your-server.example.com/mcp"
}
}
}
For individual use, each team member runs their own local instance with their own tasks.
npm run seed
npm run setup:mcp
npm run verify ```
Server runs at http://localhost:3456
Note: The data/progress.json file is gitignored - each user maintains their own task database. Run the seed script to create your initial database.
---
Here's a quick reference for all configuration file locations:
| Tool | MCP Config Location | Agent Rules Location |
|---|---|---|
| **Cursor** | ~/.cursor/mcp.json | ~/.cursor/rules/priority-forge.mdc (user-global, alwaysApply: true) or .cursor/rules/*.mdc (per-project) |
| **Droid** | ~/.factory/mcp.json | ~/.factory/AGENTS.md |
| **Claude Code** | Project root .mcp.json (shared) or ~/.claude.json via claude mcp add --scope local | ~/.claude/CLAUDE.md |
| **Codex** | ~/.codex/config.toml | ~/.codex/AGENTS.md |
Claude Code note: Claude Code readsCLAUDE.md, notAGENTS.md. Project-scoped.mcp.jsonservers prompt once for approval on first use; local-scoped servers added withclaude mcp add --scope localare the personal no-prompt option.
<details> <summary>Manual Agent Rules Configuration (click to expand)</summary>
Copy the contents of AGENT_RULES.md to the appropriate agent rules location above.
</details>
| Variable | Default | Description |
|---|---|---|
PORT | 3456 | Server port |
| Tool | Config Location | Format | Notes |
|---|---|---|---|
| **Cursor** | ~/.cursor/mcp.json | { url } | No type field needed for HTTP |
| **Droid** | ~/.factory/mcp.json | { type, url } | Requires "type": "http" |
| **Claude Code** | ~/.claude/mcp.json | { type, url } | Requires "type": "http" |
Droid and Claude Code require the explicit "type": "http" field because they support multiple transport types (HTTP, stdio, etc.). Cursor infers HTTP from the URL.
| Method | Endpoint | Description |
|---|---|---|
| GET | /health | Health check (includes version) |
| GET | /status | Full system status with top priority |
| Method | Endpoint | Description |
|---|---|---|
| GET | /top-priority | Single highest priority task |
| POST | /recalculate | Recalculate all priority scores |
| GET | /heuristic-weights | View weight configuration |
| PUT | /heuristic-weights | Update weights & recalculate |
| Method | Endpoint | Description |
|---|---|---|
GET | /workspaces | List all workspaces |
GET | /workspaces/current | Get current workspace |
POST | /workspaces | Create workspace |
POST | /workspaces/:id/switch | Switch to workspace |
DELETE | /workspaces/:id | Delete workspace |
curl http://localhost:3456/ml/export
| Prompt | Purpose |
|---|---|
start_session | Initialize work with proper task tracking |
complete_work | Close out task with completion tracking |
switch_context | Properly handle task switching |
If you used ./setup.sh or npm run setup:mcp, MCP is already configured! Otherwise, configure manually:
<details> <summary>Manual MCP Configuration (click to expand)</summary>
| Method | Description |
|---|---|
prompts/list | List available prompts |
prompts/get | Get prompt with arguments |
| Per-Workspace | Global (Shared) |
|---|---|
| Tasks | Heuristic weights |
| Projects | Completion records |
| Decisions | Selection events |
| Data gaps | Drag reorder events |
| Objectives | Online learner state |
If you started the backend with npm run dev inside an editor's integrated terminal (VS Code, Cursor, etc.), the node process becomes a child of the editor's shell. When the editor closes or crashes, it sends SIGHUP to its process tree and the backend dies with it. The frontend may appear to survive if it was started a different way (e.g. via ./setup.sh start, which uses nohup to detach).
Fix — pick one:
./setup.sh install-systemd on Linux, or re-run ./setup.sh install on macOS and accept the launchd prompt. Services run independently of any shell and auto-start on boot../setup.sh start — uses nohup so processes survive the terminal closing. Won't survive a reboot.Avoid: npm run dev in a VS Code/Cursor integrated terminal for anything other than a quick one-off test.
Priority Forge 是一个实时优先级队列系统,支持任务详细信息、工作区切换和启发式权重调节。
Priority Forge 提供以下功能: * V2 堆优先级队列:加权优先级评分与可调启发式 * V2.1 通用覆盖:MCP 资源和提示与任何 MCP 客户端(Cursor、Droid、Claude 桌面、Claude 代码)兼容 * 优先级因素:阻塞计数、跨项目影响、时间敏感性、努力/价值比率、依赖深度 * MCP 协议:JSON-RPC 2.0 端点用于 AI 助手集成
环境依赖与系统要求: * Node.js 20+ * TypeScript 5.3 * WSL2 + Ubuntu(Windows 端)
安装步骤: * 使用 `bash setup.sh` 脚本进行一键安装(推荐) * 手动安装:使用 `npm install` 和 `npm run setup:mcp` 等命令
使用教程: * 初始化数据库:`npm run seed` * 配置 MCP:`npm run setup:mcp` * 验证功能:`npm run verify`
配置说明: * MCP 配置位置:`~/.cursor/mcp.json`、`~/.factory/mcp.json` 等 * 代理规则位置:`~/.cursor/rules/priority-forge.mdc`、`.cursor/rules/*.mdc` 等
API/接口说明: * Core Endpoints:健康检查、系统状态等 * V2 Priority Endpoints:单个最高优先级任务、重新计算优先级等
工作流 / 模块说明: * MCP 提示(工作流模板):`start_session`、`complete_work`、`switch_context` 等
FAQ 摘要: * 常见问题解答
创新的AI任务编排方案���MCP生态中差异化产品。但项目初期阶段,社区认可度低,建议关注后续发展和更新频率。
AI Skill Hub 为第三方内容聚合平台,本页面信息基于公开数据整理,不对工具功能和质量作任何法律背书。
建议在沙箱或测试环境中充分验证后,再部署至生产环境,并做好必要的安全评估。
⚠️ GPL 3.0 — 强 Copyleft,衍生作品须开源,含专利保护条款,不可闭源使用。
总体来看,优先级锻造 是一款质量良好的MCP工具,在同类工具中具备一定竞争力。AI Skill Hub 将持续追踪其更新动态,建议收藏备用,结合自身场景选择合适时机引入使用。
| 原始名称 | priority-forge |
| 原始描述 | 开源MCP工具:A badass AI-powered task orchestration server. Syncs your priority queue across 。⭐7 · TypeScript |
| Topics | 任务编排优先级管理MCP工具AI调度开源 |
| GitHub | https://github.com/Unobtainiumrock/priority-forge |
| License | GPL-3.0 |
| 语言 | TypeScript |
收录时间:2026-05-20 · 更新时间:2026-05-22 · License:GPL-3.0 · AI Skill Hub 不对第三方内容的准确性作法律背书。
选择 Agent 类型,复制安装指令后粘贴到对应客户端