hypertool-mcp
Proxies to all of a user's MCPs and allows them to expose dynamic toolsets of tools from those servers to improve context and tool-usage.
README
🛠️ Hypertool MCP
Too many MCP servers? Too many tools? Poor LLM performance?
Hypertool creates dynamic toolsets that dramatically improve tool usage performance.
🎯 Why Hypertool?
The Problem: LLMs struggle when overwhelmed with too many tools. 50+ tools from multiple MCP servers? Your AI assistant becomes confused and ineffective.
The Solution: Hypertool lets you create focused, dynamic toolsets that expose only the tools you need for specific tasks. Think of it as "tool playlists" for your AI.
Only one MCP in mcp.json: Hypertool will read your mcp configuration and run as both a server (to cursor/claude code, etc.) and run or connect to your MCPs regardless of transport.
Cursor / Claude Code → Hypertool MCP → Your MCPs
📦 Multiple MCP Servers → 🎯 Focused Toolsets → 🚀 Better Performance
✨ Key Features
- 🎯 Dynamic Toolsets: Create custom tool collections from any MCP servers
- 🔄 Hot-Swapping: Switch between toolsets instantly - clients are notified automatically using MCP list tools changed notifications.
- 🌐 Universal Compatibility: Works with any MCP client (Claude Desktop, etc.)
- 🛡️ Tool Checks: Toolsets capture a hashed tool reference to validate that the right tool gets exposed.
- 📝 Tool Annotations: Add contextual notes and usage guidance to tools to help AI understand when and how to use them effectively
- 📡 Multiple Transports: Supports both stdio and HTTP/SSE protocols
🚀 Quick Start
Install hypertool in all your agentic apps.
npx -y @toolprint/hypertool-mcp@latest --install
# use the --dry-run flag to see what will be installed.
Just restart Cursor/Claude Code and it will pick up hypertool. All your MCP configs get automatically backed up so that you can restore them at any time.
Add Hypertool to your project
Claude Code
Option 1 - quick add
# In your project directory:
npx -y @toolprint/hypertool-mcp@latest --install claude-code
Option 2 - manual add
- Copy your
.mcp.jsonto.mcp.hypertool.json - Update your
.mcp.jsonwith only the hypertool MCP:
{
"mcpServers": {
"hypertool": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"@toolprint/hypertool-mcp@latest",
"--mcp-config",
".mcp.hypertool.json"
]
}
}
}
Claude Code Dynamic Toolset Support - Currently CC doesn't support MCP tool change notifications (we know, it surprised us too) but we suspect this is coming soon. We've opened an issue here. Please give it an upvote if you want this too.
In the meantime, just restart your claude code session to pick up your newly equipped toolset OR run your hypertool-mcp with the
--equip-toolset {name}flag and it will autoequip on boot.
🎯 Cursor
Option 1 - One command install
npx -y @toolprint/hypertool-mcp@latest --install cursor
This will:
- Back up your current Cursor MCP configuration
- Copy all existing servers to Hypertool's config
- Add Hypertool as your main MCP server
Option 2
Click the badge above to automatically install Hypertool MCP in Cursor IDE. This will add the server to your Cursor configuration and you can start using it immediately.
Note you will need to update the runtime flag for the hypertool mcp server --mcp-config to point to a copied version of Cursor's MCP settings. All of this is done automatically in Option 1.
Other Projects
- Copy your
mcp.jsonto.mcp.hypertool.json - Update your
mcp.jsonwith only the hypertool MCP:
{
"mcpServers": {
"hypertool": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"@toolprint/hypertool-mcp@latest",
"--mcp-config",
".mcp.hypertool.json"
]
}
}
}
🎭 Creating Your First Toolset
Once Hypertool is running, you can use the built-in MCP tools to manage toolsets:
- List available tools: Use
list-available-toolsto see what's discovered - Build a toolset: Use
build-toolsetto create a focused collection - Equip the toolset: Use
equip-toolsetto activate it
Your MCP client (like Claude) can now call these tools directly:
Claude: Use the build-toolset tool to create a development toolset with echo and add_numbers
Result: A focused toolset with only the tools you need!
🔄 Hot-Swapping Toolsets Example
Just chat with your assistant to swap toolsets using the equip-toolset tool.
Hypertool will equip the tools and send a change notification so that your assistant can read and use the newly executable tools.
Add Notes to tools!
Sometimes descriptions and reading input parameters isn't enough for an LLM to figure out how to use a tool.
Hypertool gives your agents a add-tool-annotation tool that lets them add notes to reflect on better usage the next time so that they don't need to keep spinning in circles in new contexts.
🌐 Multiple Transport Support
Stdio (Default)
Perfect for Claude Desktop and most MCP clients:
hypertool-mcp --mcp-config {path_to_config}
HTTP/SSE
Have it run separately and listen on a port over HTTP:
hypertool-mcp --transport http --port 3000 --config test-mcp.json
Access via: http://localhost:3000/mcp
🛡️ Security & Validation
Hypertool validates tool references using cryptographic hashes:
{
"name": "secure-toolset",
"tools": [
{
"namespacedName": "everything.echo",
"refId": "sha256:abc123...",
"server": "everything"
}
]
}
- Secure by default: Rejects tools with changed schemas
- Insecure mode: Available with
--insecureflag (use carefully!) - Automatic validation: Toolsets are re-validated when servers reconnect
📋 Common Workflows
Development Setup
1. Start Hypertool: hypertool-mcp --config test-mcp.json
2. Ask Claude: "Use build-toolset to create a frontend toolset with echo and add_numbers"
3. Ask Claude: "Use equip-toolset to activate the frontend toolset"
Production Debugging
1. Ask Claude: "Use equip-toolset to switch to debug toolset"
2. Your AI now has access to focused debugging tools
3. MCP clients are automatically notified of the change
Team Collaboration
1. Ask Claude: "Use list-saved-toolsets to see available toolsets"
2. Share toolset configurations via your preferred method
3. Team members can equip the same toolsets for consistency
🔧 Configuration Reference
CLI Options
hypertool-mcp [options]
Options:
--config <path> MCP servers config file (default: .mcp.json)
--transport <type> Transport type: stdio|http (default: stdio)
--port <number> HTTP port (default: 3000)
--equip-toolset <name> Load toolset on startup
--debug Verbose logging
Server Configuration
{
"mcpServers": {
"server-name": {
"type": "stdio|sse",
"command": "command-name",
"args": ["arg1", "arg2"],
"env": {
"ENV_VAR": "value"
}
}
}
}
🤝 Integration Examples
Equip a specific toolset on server spawn
{
"mcpServers": {
"hypertool": {
"type": "stdio",
"command": "hypertool-mcp",
"args": ["--config", "/path/to/test-mcp.json", "--equip-toolset", "daily-dev"]
}
}
}
📊 Performance Impact
Before Hypertool:
- 🐌 50+ tools exposed to LLM
- 😵 Confused tool selection
- 🔄 Slow response times
After Hypertool:
- ⚡ 5-10 focused tools per context
- 🎯 Precise tool selection
- 🚀 Faster, more accurate responses
🛠️ Development
# Clone and setup
git clone https://github.com/yourorg/hypertool-mcp
cd hypertool-mcp
npm install
# Development
npm run dev # Start with hot reload
npm test # Run tests
npm run lint # Code quality
📄 License
MIT License - see LICENSE file for details.
<div align="center">
Built by developers who got tired of overwhelming their AI with too many tools. Hope you find it useful! 🚀
<a href="https://toolprint.dev"> <img src="./assets/toolprint.png" alt="Toolprint" width="200"> </a>
<p> <strong>Built with ❤️ by <a href="https://toolprint.dev">Toolprint</a></strong><br> <sub>© 2025 OneGrep, Inc.</sub> </p>
</div>
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。
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。
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。
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。