
Mcp Launcher
io.github.oneprofile-dev由 GitHub 账号 oneprofile-dev 发布v2.1.1更新于 Oct 24, 2026
Single MCP hub: discover, add, audit, and govern MCP servers across every AI client.
概览
一个本地 MCP 中枢,通过单一入口在多个 AI 客户端中发现、添加、审计和治理其他 MCP 服务器。
- 可以做什么
- 以单个 stdio MCP 中枢运行,向外分发到你添加的每个服务器,因此在每个 AI 客户端中只需一份配置即可暴露全部服务器,工具名带 slug 前缀。提供 search_servers、get_server_details、add_to_stack 等发现工具,以及审计 MCP 配置风险、添加或移除服务器、列出当前栈、打印客户端配置片段的命令行命令。guard 模式用本地策略引擎包裹另一个 MCP 服务器并拦截工具调用,login 与 sync 可拉取团队注册表配置并推送审计结果。
- 适用场景
- 适合在多个 AI 客户端中使用 MCP 服务器、希望只配置一次而不必反复粘贴配置的场景。也适合检查机器上已有哪些 MCP 服务器,以及在团队中统一使用经过审核的服务器列表。
- 运行要求
- 通过 npx 启动的本地进程,需要 Node.js 18 或更高版本。仅支持桌面客户端,无网页可执行版本。审计与中枢使用无需认证;login 需要 CuratedMCP 账户的注册表密钥。可选环境变量 CURATOR_API_URL 可覆盖 API 基础地址,CURATOR_TELEMETRY=false 可关闭遥测。
安装
在 SourceWeft 中
- 打开 控制台中的 Mcp Launcher,将其添加到工作区。
- 为需要使用其工具的对话启用该服务。
Desktop only,通过 STDIO。 STDIO 服务会启动本地进程,因此需要 SourceWeft 桌面宿主。
其他 MCP 客户端
参照 仓库 中的启动说明。
README
curatedmcp
[npm version] [npm downloads] [License: MIT] [Node.js ≥18]
The CuratedMCP Agent. One CLI to discover, run, audit, and govern every MCP server your AI tools (Claude, Cursor, Windsurf, Copilot, Gemini) use.
Plug it in once. Add servers anytime. Audit and govern them from one place.
What you get
1. Audit (the wedge — start here)
Scans every MCP config file on your machine (Claude Desktop, Cursor, Windsurf, Claude Code, …), classifies each server against the CuratedMCP catalog, and flags:
- 🔴 HIGH — unverified or known-risky servers with credentials
- 🟡 MEDIUM — verified servers running outside catalog defaults
- 🟢 VERIFIED — known-good catalog servers
No signup, no cloud, no data leaves your machine. Logged in? Add --sync to push the result to your dashboard.
2. Run as the MCP Hub
If you use MCP servers across multiple AI clients, you've felt this pain: configure GitHub MCP in Claude Desktop, then re-do it in Cursor, then in Windsurf. New agent ships? Re-paste every config.
The agent fixes that. It's one MCP entry that fans out to every server you've added, in every AI client.
Add it to your AI client
Add servers to your stack
Restart your AI client
Tools appear with a <slug>__ prefix:
github__create_issuepostgres__queryfilesystem__read_file
3. Guard (local action firewall)
Wraps an MCP server with a local policy engine that gates every tools/call against
~/.curatedmcp/guard-policy.json. Default policy allows read, prompts on write, blocks destructive.
4. Login + sync (for teams)
Once you have a CuratedMCP account, link the CLI to it:
Sync pulls the locked-down server list approved by your team and merges it into your local stack — so every developer's machine runs the same vetted set of servers.
Config files
~/.curatedmcp/stack.json — your stack, plain JSON, hand-editable, version-controllable:
Set "disabled": true on an entry to skip it without removing it.
Other files (created on first use):
~/.curatedmcp/auth.json— login token (mode 0600)~/.curatedmcp/guard-policy.json— firewall policy~/.curatedmcp/launcher.json— anonymous client UUID
In-agent discovery
The agent itself exposes discovery tools to your AI client, so you can ask:
"Find me an MCP server for Postgres." "What's the best Stripe MCP?" "Add the Postgres MCP server to my stack."
The agent uses search_servers, get_server_details, and add_to_stack to do all of that without you leaving the chat.
Privacy
- All config is local at
~/.curatedmcp/. No cloud sync unless youlogin. - Anonymous telemetry only (event names like "search", "add"). Disable with
--no-telemetryorCURATOR_TELEMETRY=false. - Audit results stay on your machine unless you
loginand run--sync.
Compatibility
- Works with Claude Desktop, Claude Code, Cursor, Windsurf, Copilot, Gemini, OpenAI Agents — anything that supports MCP over stdio.
- Node.js ≥ 18.
Migrating from the old packages
The agent replaces three earlier packages, which are now deprecated:
A launcher bin alias is kept for back-compat.
Links
MIT licensed.
来源:README.md,提交 80f4d67
工具
0版本历史
1- v2.1.1最新Sep 16, 2026
评分
已满足适用检查项的 50%
该评分衡量此收录条目的完整程度,依据注册中心条目、端点与公开来源计算得出。它不是对服务器本身的评价,也不表示服务器安全或当前可用:可连通性是带日期的检查结果,评分也不包含安装量或用户评价。
- 近 30 天内可连通25 分
请保持端点能够响应 MCP 握手;SourceWeft 每 30 天探测一次。
- 已声明认证方式10 分
请在注册中心条目中声明凭据请求头,或保持端点能够响应探测,以便判定其认证方式。
- 已列出工具15 分
请让服务器在无凭据时响应 tools/list,或发布可在沙箱中无凭据启动的软件包。
- 描述必需15 分
- 图标10 分
- 文档15 分
- 许可证10 分