AIForge
Unified registry & router for agent skills, MCP servers, and Claude Code plugins.
Install / Use
claude mcp add sheep-programmer -- npx -y github:sheep-programmer/AIForgeIf the server publishes to npm under a different name, use that package instead — check the repo README.
MCP Server
Model Context Protocol server
Quality Score
Category
Development & EngineeringSupported Platforms
Our assessment of AIForge
AIForge scores 74/100 on our quality scale, 3791st of 4,698 Development & Engineering skills we index.
Its MCP Server is 9.4 KB long, well organised into 15 sections with 9 code examples: a thorough specification that gives an agent plenty to work with.
It has 3 GitHub stars, so there is little community track record yet; judge it on its content.
Maintenance, license and trust
- The repository was last updated about 5 months ago. That is recent enough to be usable, but agent tooling moves fast, so check the instructions against your agent's current version.
- Our last check on 2026-11-12 found the source still online.
- No license is declared. By default that means all rights are reserved: you can read it, but reusing or redistributing it is not clearly permitted. Ask the author before building on it commercially.
- Its trust signals score 73/100, with 3 cautions from licensing, adoption, age or documentation. These come from repository metadata, not a code audit — read the skill file before letting an agent act on it.
Safety scan
No issues foundOur scan of the whole file found no instruction hijacking, hidden characters, credential access, data exfiltration or destructive commands. An AI review of the same text found nothing harmful.
AI review by kimi-k2.7-code on 2026-09-24. Automated pattern scan on 2026-09-24. It catches known dangerous patterns, not every risk — read a skill before letting an agent act on it.
AIForge compared with similar skills
All 4 of these similar skills score higher than AIForge; compare them before choosing.
| Skill | Score | Stars | Updated | Format |
|---|---|---|---|---|
| AIForge (this skill)by sheep-programmer | 74 | 3 | 5mo ago | MCP Server |
| taste-skillby Leonxlnx | 100 | 116.8k | 8d ago | CLAUDE.md |
| Agent-Reachby Panniantong | 100 | 112.3k | 8d ago | CLAUDE.md |
| rufloby ruvnet | 100 | 81.9k | today | CLAUDE.md |
| headroomby headroomlabs-ai | 100 | 79.8k | today | CLAUDE.md |
Frequently asked questions
- How do I install AIForge?
- Run
claude mcp add sheep-programmer -- npx -y github:sheep-programmer/AIForge. The install tabs above show the steps for each supported agent. - Which AI agents does AIForge work with?
- It is written for Claude Code and Claude Desktop, as a MCP Server file. Other agents that read the same format can often use it too.
- Is AIForge safe to use?
- Our scan of the whole file found no instruction hijacking, hidden characters, credential access, data exfiltration or destructive commands. An AI review of the same text found nothing harmful. It declares no license and scores 73/100 on trust signals. Skills are instructions an agent will follow, so read the file before installing it and do not approve commands you do not understand.
- Is AIForge still maintained?
- The repository was last updated about 5 months ago. That is recent enough to be usable, but agent tooling moves fast, so check the instructions against your agent's current version.
Skill content
View source on GitHub<sub>Skill / MCP / Plugin 一站式注册中心 · 由小模型按需挑选 · 只把真正需要的几个注入对话上下文</sub>
</div>✦ 为什么需要 AIForge
2026 年,Claude Code / Codex / Cursor 的扩展生态彻底爆炸:
anthropics/skills、obra/superpowers、vercel-labs/skills、pbakaus/impeccable—— 公开 skill 过万@modelcontextprotocol/server-*、@playwright/mcp、各类自建 MCP server —— 暴露成百上千个 tool- Claude Code plugin 市场刚起步但已经在分裂
三个东西凑在一起,三个老大难问题:
| | 问题 | AIForge 的解法 | |---|---|---| | ① | Token 浪费 —— 200 个 skill + 30 个 MCP 的 tool 列表塞进 context,还没开聊就烧掉一半预算 | 服务端跑小模型,每次提问只挑 top-N (默认 3) 注入 | | ② | 功能冲突 —— 三个 review skill / 五个 browser-automation MCP;agent 要么乱挑要么全跑 | 向量去重 + 小模型重排,每簇只保留最优代表 | | ③ | 管理割裂 —— skill 在 git、MCP 在 settings.json、plugin 在 ~/.claude/plugins/ | 统一 Artifact 模型 + 一个 Web 面板掌管 |
✦ Web 管理面板
<div align="center"> <a href="docs/web-admin.md"> <img src="docs/assets/web-screenshot.png" alt="AIForge Web 管理面板截图" width="92%"/> </a> <br/> <sub>「Editorial Engineering」设计语言 · 暖色 parchment 底 + 单一氧化铜绿强调 · Fraunces 衬线标题 + Inter 正文 + JetBrains Mono 数据</sub> </div> <br/>十一条路由,全部中文,企业级密度:
| 路径 | 内容 |
|---|---|
| / | Dashboard · KPI · 24h 流量曲线 · 最近活跃 · 新手 4 步引导 |
| /artifacts | 浏览全部 artifact,按类型 / 标签 / 仓库筛选;URL state 可分享 |
| /artifacts/[id] | 详情:body、metadata、tag 编辑器、mcp_config 一键复制、plugin manifest |
| /tags | 20 个预置标签 + 自定义;artifact 使用度可视化 |
| /ingest | 粘贴 GitHub URL → 实时状态机时间线 |
| /autotag | 小模型批量打标 + 进度条 + ETA + 实时操作流 |
| /playground | 输入 prompt 看 top-K 推荐 + score bar + rerank 理由 |
| /insights | 召回质量 / 延迟分布 / 覆盖矩阵 / 流量热力的深度分析面板 |
| /discovery | 远程 finder 找到的高质量仓库审批队列 |
| /environment | 本机各家 agent(Claude Code / Codex / Cursor / Gemini / Windsurf)已装的 MCP / plugin / skill 清单 |
| /settings | API 地址 / API Key / 默认 top-K / 主题 |
<sub>真实截图 · 演示数据(后端未连接时前端自动回落,UI 始终可演示)</sub>
<table> <tr> <td width="33%"><a href="docs/assets/screen-insights.png"><img src="docs/assets/screen-insights.png" alt="Insights 深度分析面板"/></a><br/><sub><b><code>/insights</code></b> · 召回质量 / 延迟分布 / 覆盖矩阵</sub></td> <td width="33%"><a href="docs/assets/screen-playground.png"><img src="docs/assets/screen-playground.png" alt="Playground 推荐试验"/></a><br/><sub><b><code>/playground</code></b> · prompt → top-K 推荐 + 理由</sub></td> <td width="33%"><a href="docs/assets/screen-environment.png"><img src="docs/assets/screen-environment.png" alt="本机环境扫描"/></a><br/><sub><b><code>/environment</code></b> · 跨 agent 已装清单</sub></td> </tr> </table>cd web
npm install
npm run dev # 默认 http://localhost:3500
后端不可达时,前端自动 fallback 到 demo 数据,UI 永远可演示。
✦ 工作原理
flowchart LR
CC["Claude Code<br/>(plugin / hook)"] -- HTTP /v1/recommend --> API[FastAPI]
CC -- inject top-N --> CTX[Context Window]
Web["Web Admin<br/>(Next.js)"] <-- HTTP --> API
GW["aiforge-mcp<br/>runtime gateway"] -- JSON-RPC stdio --> MCPs[(N × downstream MCP)]
GW -- HTTP active set --> API
CC <-- MCP stdio --> GW
subgraph SRV["AIForge server"]
API --> REC[recommender]
API --> ING[ingestion]
REC -- vector search --> DB[(SQLite + sqlite-vss)]
REC -- rerank --> LLM[Qwen-1.5B / Haiku]
ING --> DB
TAGGER[auto-tagger] --> DB
end
服务端不可达时,插件自动切到本地兜底模式:缓存的 SQLite + 关键词索引让它依然能挑出合理的 artifact,只是少了小模型重排。
✦ 快速上手
1. 启动服务端
git clone https://github.com/sheep-programmer/AIForge.git aiforge
cd aiforge/server
docker compose -f docker/docker-compose.yml up -d
# HTTP API → http://localhost:8765
种子入库:
curl -X POST http://localhost:8765/v1/ingest \
-H 'Content-Type: application/json' \
-d '{"github_url": "https://github.com/obra/superpowers-skills"}'
curl -X POST http://localhost:8765/v1/ingest \
-H 'Content-Type: application/json' \
-d '{"github_url": "https://github.com/anthropics/skills"}'
入库时自动检测仓库内容 —— .claude-plugin/plugin.json 进 plugin 表、mcp.json 进 mcp 表、SKILL.md 进 skill 表。一个仓库可以同时产出多种 artifact。
2. 装插件(自动注入推荐)
cd aiforge/plugin
./install.sh --server http://localhost:8765
在 ~/.claude/settings.json 写入一个 UserPromptSubmit hook;plugin 落到 ~/.claude/plugins/aiforge/。
3. 启 Web 管理面板(可选但推荐)
cd aiforge/web
npm install
npm run dev # → http://localhost:3500
4. (可选)启 MCP 运行时网关
让 Claude Code 只对接一个 MCP,由 AIForge 路由到 N 个下游:
// ~/.claude/settings.json
{
"mcpServers": {
"aiforge": { "command": "aiforge-mcp" }
}
}
aiforge-mcp 启动时拉取 active MCP 集合,聚合 tool 列表,按 <name>__<tool> 命名空间路由。
✦ 核心特性
<table> <tr> <td width="50%" valign="top">🗂 统一 Artifact 模型
一张 Skill 表通过 artifact_type 字段同时承载 skill / mcp / plugin;统一 /v1/artifacts API。
🏷 扁平多标签 + 自动打标 20 预置 tag + 任意自定义。小模型复用 reranker 给每个 artifact 挑 1-3 个最贴合的 tag。
</td> </tr> <tr> <td valign="top">⚡ 两阶段推荐器
all-MiniLM-L6-v2 向量召回 top-30 → Qwen2.5-1.5B(或 Haiku)重排成 top-3。
🔌 MCP 运行时网关
aiforge-mcp 进程对外是单个 MCP,对内连 N 个下游,按 <name>__<tool> 命名空间路由。
♻️ 自动去重 把语义等价的 artifact 聚类,按来源信誉、更新时间、安装量挑最优代表。
</td> <td valign="top">🛠 一键安装
Web 面板或 /aiforge:install 把 MCP 写进 settings.json、plugin 拉到 ~/.claude/plugins/,自动备份。
🛰 远程 skill-finder(默认关闭) 定期扫描 GitHub 高质量新仓库,进人工审批队列。
</td> <td valign="top">🪂 本地兜底 服务端挂了,插件用缓存索引继续工作。
</td> </tr> </table>✦ Slash 命令
装完插件后可用:
/aiforge:status # 服务端健康 + 本地缓存状态
/aiforge:add <github-url> # 入库一个仓库
/aiforge:search <query> # 关键词搜索 artifact
/aiforge:sync # 拉取最新索引到本地缓存
/aiforge:config # 查看 / 修改插件配置
/aiforge:list [--type=...] # 列出 artifact(含已安装标记)
/aiforge:install <id> # 安装 MCP 到 settings.json / plugin 到 ~/.claude/plugins
/aiforge:uninstall <id> # 反向卸载
/aiforge:tag <id> <t1,t2> # 手动打标
/aiforge:autotag # 触发自动打标 + 实时进度
✦ 架构
详见 docs/architecture.md(中文)/ docs/architecture.en.md(English)。架构图源文件在 docs/diagrams/。
| 组件 | 技术选型 | 选型理由 |
|------|----------|----------|
| HTTP API | FastAPI + Uvicorn | 异步、自带 OpenAPI、生态成熟 |
| 向量库 | SQLite + sqlite-vss | 零运维、嵌入式、10 万级 artifact 足够快 |
| 向量编码 | sentence-transformers (all-MiniLM-L6-v2) | 384 维、CPU 可跑、benchmark 充分 |
| 重排 / 打标 | Ollama 跑 Qwen2.5-1.5B(默认) / Claude Haiku API | 体积小、速度快、排序质量出乎意料 |
| MCP 网关 | asyncio + JSON-RPC | 无额外依赖,未来可换官方 SDK |
| Web 管理面板 | Next.js 15 + React 19 + Tailwind + 自研 shadcn 风格组件 | 前端类型源自后端 OpenAPI,可部署在 FastAPI 后面 |
| 插件 | bash + Python(stdlib only) | 原生 Claude Code,零三方依赖 |
✦ 项目状态
v0.1 ▰▰▰▰▰▰▰▰▰▰ shipped · server + recommender + ingest + plugin + fallback + remote-finder
v0.2 ▰▰▰▰▰▰▰▰▰▰ shipped · unified artifact · tags · autotag · web admin · MCP gateway MVP
v0.3 ▱▱▱▱▱▱▱▱▱▱ in design · gateway hot-reload · cross-agent (Codex/Cursor/Gemini) · web i18n
完整 roadmap:docs/roadmap.md
✦ 贡献
见 CONTRIBUTING.md。当下最需要的贡献:
- 真实世界中的 artifact 库样本(用来测试推荐 / 打标质量)
- 去重测试用例
- 重排器 / 打标器 prompt 优化
- Web 面板 i18n(English / 日本語 / Deutsch)
✦ 许可
Apache 2.0 —— 见 LICENSE。
Related Skills
taste-skill
116.8kAgent skills that give coding agents taste: art direction, reference images and a browser check instead of generic AI pages
Agent-Reach
112.3kGive your AI agent eyes to see the entire internet. Read & search Twitter, Reddit, YouTube, GitHub, Bilibili, XiaoHongShu — one CLI, zero API fees.
ruflo
81.9k🌊 The original agent harness. Deploy intelligent multi-player swarms, coordinate autonomous workflows, and build conversational AI systems. Features adaptive memory, self-learning intelligence, federation, vector RAG integration, and native Claude Code / Codex / Hermes and many more Integrated
headroom
79.8kCompress tool outputs, logs, files, and RAG chunks before they reach the LLM. 20% fewer tokens for coding agents, 60-95% fewer tokens for JSON, same answers. Library, proxy, MCP server.
Languages
Trust signals
From repository metadata: license, adoption, age and documentation. Not a code audit — see the Safety scan above for what the skill file itself contains.
