经 AI Skill Hub 精选评估,stashbase 获评「推荐使用」。这款MCP工具在功能完整性、社区活跃度和易用性方面表现出色,AI 评分 7.5 分,适合有一定技术背景的用户使用。
stashbase 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。
stashbase 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/liliu-z/stashbase
# 方式二:手动配置 claude_desktop_config.json
{
"mcpServers": {
"stashbase": {
"command": "npx",
"args": ["-y", "stashbase"]
}
}
}
# 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 stashbase 执行以下任务... Claude: [自动调用 stashbase 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": {
"stashbase": {
"command": "npx",
"args": ["-y", "stashbase"],
"env": {
// "API_KEY": "your-api-key-here"
}
}
}
}
// 保存后重启 Claude Desktop 生效
Stash your knowledge. Agents maintain it.
⭐ If this idea interests you, drop a star. Every item on the build map ships with detailed design — easy to pick up if you'd like to contribute.
StashBase is an agent-native, local-first knowledge base — built for personal context that compounds over time and stays portable across AI tools.
🤖 Agent-native: Agents organize and maintain the KB per rules in STASHBASE.md. The same KB is exposed via MCP to both built-in Claude Code / Codex and any external MCP-compatible client — Claude Desktop, ChatGPT Desktop, and more.
💾 Local-first & user-owned: Plain HTML / Markdown / PDF files on your disk, fully portable. The accumulated context is your asset.
🎨 Native HTML: Native HTML rendering inside the app, plus a PDF → HTML pipeline that turns fixed-layout, read-only PDFs into pages agents can edit, maintain, and retrieve.
---
Built by Li Liu.
I work on Milvus at Zilliz, where I've spent the last few years building vector retrieval infrastructure for AI systems.
Coding with agents already feels fluid inside IDEs. Personal knowledge tools still largely don't.
StashBase is my attempt at the missing layer: an agent-native, local-first workspace where papers, notes, transcripts, and saved analysis remain continuously retrievable across AI workflows.
This is a personal side project built in the open. PRs, issues, and experiments are welcome.
For contributors and developers building locally, and for platforms without a prebuilt cask (Intel Mac, Windows, Linux). End users should wait for the brew cask above.
```bash
cd StashBase pnpm install pnpm setup:python
pnpm dist:mac pnpm dist:win ```
Most AI tools treat knowledge as temporary context — uploaded files, copied notes, chat history. But personal knowledge accumulates over years: papers, transcripts, notes, drafts, fragments spread across folders and apps.
StashBase is built for that accumulation layer. Files live as plain HTML / Markdown / PDF on your disk; a local indexer keeps them retrievable through hybrid (semantic + keyword) search; the whole KB is exposed via MCP to any AI client.
research paper
↓
Claude Code reads and generates an HTML note
↓
StashBase indexes it locally
↓
later, in Claude or ChatGPT...
↓
"that paper on test-time compute scaling"
↓
relevant notes retrieved instantly
Codebase retrieval is already well served by tools like Claude Context. StashBase focuses on everything else that accumulates around AI work — papers, transcripts, reading notes, research fragments, saved analysis, generated HTML knowledge pages.
---

Import the CS183B starter — 20 YC startup lectures, pre-indexed. Surface ideas in Claude Desktop via @stashbase, then use the built-in agent (Claude Code) to organize what resonates into HTML notes.
---
Drop a PDF in and a converter generates a readable, indexable HTML companion (paper.html + paper_files/ asset bundle). Original PDF is kept alongside. Failed conversions surface a Retry button — never auto-retried (failures are usually persistent: scanned-only, encrypted, etc.).
The packaged app ships an MCP command for @stashbase. Open Settings → MCP (gear in the top-right corner) to connect it to Claude Code, OpenAI Codex CLI, Gemini CLI, Qwen Code, Cursor, Void, Claude Desktop, Windsurf, VS Code, Cherry Studio, Cline, Augment, Roo Code, Zencoder, ChatGPT Desktop, LangChain/LangGraph, or any other MCP-aware client.
Use the manual config below if you're running from source, if a client was installed after StashBase, or if you want to inspect the exact MCP settings.
Connector support. From Settings → MCP, StashBase writes the right config file directly for Claude Code, Claude Desktop, OpenAI Codex CLI, Gemini CLI, Qwen Code, and Cursor. For GUI-driven or extension-managed clients (Void, Windsurf, VS Code, Cherry Studio, Cline, Augment, Roo Code, Zencoder, ChatGPT Desktop, LangChain/LangGraph, …) the same Connector button copies the stdio config to paste into the client's MCP settings.
stashbase是一个有潜力的开源MCP工具,值得关注
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。
✅ Apache 2.0 — 宽松开源协议,可商用,需保留版权声明和 NOTICE 文件,含专利授权条款。
AI Skill Hub 点评:stashbase 的核心功能完整,质量良好。对于Claude Desktop / Claude Code 用户来说,这是一个值得纳入个人工具库的选择。建议先在非生产环境试用,再逐步推广。
| 原始名称 | stashbase |
| 原始描述 | 开源MCP工具:Local knowledge base for AI workflows. HTML-first, continuously indexed, MCP-com。⭐20 · TypeScript |
| Topics | mcptypescriptai |
| GitHub | https://github.com/liliu-z/stashbase |
| License | Apache-2.0 |
| 语言 | TypeScript |
收录时间:2026-06-02 · 更新时间:2026-06-02 · License:Apache-2.0 · 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