Notes — local markdown knowledge base
Local markdown notes for any MCP client: ranked + semantic search, tags, todos, a wiki-link knowledge graph, daily notes, templates, and slash-command workflows. Pure-JS, local, no API keys.
README
<div align="center">
🧰 mcp-suite
A collection of Model Context Protocol servers, built in TypeScript.
</div>
Each server lives in its own folder under servers/ and publishes to npm as an independent package, while sharing tooling through an npm workspace. Connect them to any MCP client — Claude Desktop, Claude Code, Cursor, and more.
Table of contents
- Servers
- Quickstart
- Install
- Connecting to a client
- Architecture & conventions
- Development
- Adding a new server
- Publishing
- Contributing
- License
Servers
| Server | Description | Status |
|---|---|---|
notes |
Local markdown notes: ranked full-text and semantic search, tags, todos, a [[wiki-link]] knowledge graph, daily notes + templates, and slash-command workflows (prompts) |
✅ Stable |
files |
Sandboxed local filesystem: read, glob + content search, token-efficient edits, copy/move, soft-delete trash, zip, checksums, and dedup | ✅ Stable |
github |
GitHub: search repos/code/issues, read repos/issues/PRs/files, notifications, create issues — OAuth device flow or token | ✅ Stable |
git |
Git: status, log, diff, file history, branches/tags, gated stage/commit, and remote clone/fetch/pull/push — pure-JS, no git binary | ✅ Stable |
More on the way: a Spotify controller, and others.
Quickstart
git clone https://github.com/Abhishekkumar2021/mcp-suite.git
cd mcp-suite
npm install # installs deps for every workspace
npm run build # builds every server
Install
Every server ships through every common channel. For the Claude Code plugin, add the marketplace
once — /plugin marketplace add Abhishekkumar2021/mcp-suite — then install per the table. See each
server's README for full config.
| Server | Claude Code plugin | npm (any client) | Claude Desktop (MCPB) | MCP registry |
|---|---|---|---|---|
notes |
/plugin install notes |
npx -y @abhishekmcp/notes |
drag notes-*.mcpb from the latest release |
io.github.Abhishekkumar2021/notes |
files |
/plugin install files |
npx -y @abhishekmcp/files |
drag files-*.mcpb |
io.github.Abhishekkumar2021/files |
github |
/plugin install github |
npx -y @abhishekmcp/github |
— | io.github.Abhishekkumar2021/github |
git |
/plugin install git |
npx -y @abhishekmcp/git |
— | io.github.Abhishekkumar2021/git |
notesdefaults to~/notes;filesrequiresFS_ROOTS(the directories it may touch). MCPB bundles install via Claude Desktop → Settings → Extensions.
Connecting to a client
Each server's README has full configuration instructions. As an example, to use the notes server with Claude Code:
claude mcp add notes --env NOTES_DIR=$HOME/notes -- node "$(pwd)/servers/notes/dist/index.js"
Or with Claude Desktop (claude_desktop_config.json):
{
"mcpServers": {
"notes": {
"command": "node",
"args": ["/absolute/path/to/mcp-suite/servers/notes/dist/index.js"],
"env": { "NOTES_DIR": "/absolute/path/to/your/notes" }
}
}
}
Architecture & conventions
A few principles hold across every server:
- Pure-JS, no native dependencies. Servers run via
npx/MCPB on any machine, so anything needing a native build is off-limits (e.g.notesuses MiniSearch + WebAssembly embeddings instead of SQLite;filesusesfast-globinstead of ripgrep).npm auditstays clean. - Sandbox-first, layered design. Each server keeps a thin
index.ts(tool registration only) over focused modules, with a single security boundary every path must pass through (realpath-containment, symlink-escape rejection, atomic writes). - Token-efficient by default. Pagination, head/tail/section reads, snippets, and compact graph refs keep tool output small.
- Automated, hands-off releases. A GitHub Release tagged
<server>-v<version>publishes to npm (with provenance), the official MCP registry, and attaches an MCPB bundle — all via OIDC, no secrets. - Tested in CI. Each server ships a committed
node:testsuite (unit + integration/security) run on every push.
Development
This is an npm workspaces monorepo.
npm install # install all workspace dependencies (run once at the root)
npm run build # build every server
npm run build -w servers/notes # build a single server
npm test --workspaces --if-present # run each server's test suite
npm run clean # remove all build output
Adding a new server
- Create
servers/<name>/with its ownpackage.json(unique npmname) and atsconfig.jsonthatextends: "../../tsconfig.base.json". - Put source in
servers/<name>/src/. - Run
npm installat the root so the workspace picks it up. - Add a row to the Servers table above.
All servers follow the naming standard (@abhishekmcp/<slug>). See CONTRIBUTING.md for detailed guidelines.
Publishing
Releases are automated. Cutting a GitHub Release tagged <server>-v<version> triggers
publish.yml, which — authenticated entirely via GitHub OIDC (no
NPM_TOKEN) — publishes the package to npm with provenance, registers it on the official MCP
registry, and builds + attaches the MCPB bundle to the release.
# bump servers/<name>/package.json, commit, then:
gh release create <name>-v<version> --title "<name> v<version>" --notes "…"
Contributing
Contributions are welcome! Please read CONTRIBUTING.md and our Code of Conduct. Found a security issue? See SECURITY.md.
License
MIT © Abhishek
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。
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。
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。
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。