Organize and Focus AI Tools
MCP proxy that groups tools from unlimited MCP servers into focused, hot-swappable toolsets for your AI.
Maintainer of this project? Claim this page to edit the listing.
0.0.45Add to Favorites
Why it matters
Streamline AI tool selection by organizing unlimited MCP servers into focused, hot-swappable toolsets. Reduce overload and improve accuracy for specific tasks.
Outcomes
What it gets done
Create specialized tool collections (3-5 tools) for specific tasks.
Dynamically switch between toolsets without restarting.
Leverage pre-configured personas for common development roles.
Monitor server health and reconnect automatically.
Install
Add it to your toolbox
Run in your project directory:
curl -fsSL https://spark.entire.vc/get/vb-hypertool | bash Capabilities
Tools your agent gets
Shows token estimation for all available tools across all MCP servers
Shows token usage for your current active tool set
Overview
Hypertool MCP Server
An MCP proxy server that lets you connect unlimited MCP servers while exposing your AI only a focused, hot-swappable toolset at a time, with pre-configured personas and per-tool token measurement. Use once too many connected MCP servers cause wrong tool picks or hit a tool-count ceiling. Hot-swap without restart works in Cursor/VSCode; Claude Desktop needs a restart after switching toolsets.
What it does
HyperTool MCP is a local proxy server that sits between your AI (Claude, Cursor, etc.) and all your configured MCP servers, letting you connect unlimited servers - 10, 50, or 500+ tools total - while your AI only sees a focused, task-specific subset at any time. Instead of exposing all tools from every server (which causes wrong tool picks, slow decisions, and confused context once you cross roughly 50-100 tools), HyperTool lets you build named toolsets like "coding" (5 focused tools instead of 47 raw Git commands) and switch between them instantly.
Setup copies your existing MCP config to .mcp.hypertool.json, then points your AI's MCP config at HyperTool itself, which proxies through to the real servers.
{
"mcpServers": {
"hypertool": {
"command": "npx",
"args": ["-y", "@toolprint/hypertool-mcp", "mcp", "run", "--mcp-config", ".mcp.hypertool.json"]
}
}
}
Personas are pre-configured MCP server bundles with curated toolsets for common workflows - web-dev (Git, Docker, Filesystem, Browser, Testing), data-scientist (Python, Jupyter, Database, Filesystem, Plotting), devops (Docker, Kubernetes, AWS, Terraform, Monitoring), content-creator (Notion, Slack, Grammar, SEO, Social), and researcher (Perplexity, Arxiv, Wikipedia, Filesystem) - installed from a separate awesome-mcp-personas repository and runnable with --persona <name>, optionally combined with your own --mcp-config servers.
Context measurement shows token estimates for every available tool and for the currently equipped toolset via the list-available-tools and get-active-toolset MCP tools, using BPE-based approximation for consistent relative comparisons (not exact per-model token counts) with under 10ms overhead. Tool annotations let you enhance tool descriptions with examples and context to improve the AI's tool-selection accuracy. Server groups organize MCP servers into logical sets so related servers launch together per project. HTTP mode runs HyperTool as a long-lived server for persistent connections; in stdio mode, a toolset is chosen at launch with --equip-toolset <name>.
Toolsets, personas, and preferences are all stored locally under ~/.toolprint/hypertool-mcp/ and can be edited directly while HyperTool is not running.
When to use - and when NOT to
Use this connector when you've connected enough MCP servers that your AI is choosing the wrong tool or hitting a tool-count ceiling (commonly cited around 100 tools), and you want task-specific toolsets (coding, writing, debugging, research) that your AI can hot-swap between rather than seeing every tool from every server at once.
Hot-swapping without a restart is fully supported in Cursor/VSCode; Claude Desktop currently requires an app restart after switching toolsets, and Claude Code hot-swap support was still in progress as of this writing (tracked via the --equip-toolset flag). Multiple simultaneous active toolsets are supported in stdio mode via the equip-toolset flag; HTTP mode supports one active toolset at a time, switchable on demand. The hypertool-mcp service subcommand is currently disabled.
Capabilities
Dynamic toolset creation, switching, and management (build/modify/equip toolsets on the fly); persona installation and use (bundled MCP servers plus curated toolsets for a workflow); context measurement (per-tool and per-toolset token estimates via list-available-tools and get-active-toolset); tool annotation enhancement for better AI tool selection; server grouping for project-based organization; and HTTP mode for persistent, long-lived connections.
How to install
Run via npx with no separate install step, pointing your MCP client's config at HyperTool instead of your servers directly: npx -y @toolprint/hypertool-mcp mcp run --mcp-config .mcp.hypertool.json for standard mode, or npx -y @toolprint/hypertool-mcp mcp run --persona web-dev for persona mode. For local development, clone the repository and run just setup-dev (Node.js 18+ and Python 3.8+ required for pre-commit hooks).
Who it's for
Developers and teams running many MCP servers who need their AI to stay focused on a small, relevant tool set per task instead of drowning in every available tool at once.
Source README
Give your AI the best tools from all your MCPs ๐ฏ
โก Features
๐ Break Free from Tool Limits
Connect unlimited MCP servers. Use 10, 50, or 500+ tools total - your AI only sees what it needs.
๐ฏ Task-Specific Toolsets
Build "git-essentials" with 5 tools instead of drowning in 47 Git commands. Switch contexts instantly.
๐ง Smart Tool Descriptions
Enhance tools with examples and context. Watch your AI pick the right tool 89% more often.
๐ Quick Start
Step 1: Copy Your Existing Config
# In your project directory
cp .mcp.json .mcp.hypertool.json
Step 2: Point Your AI to HyperTool
Replace your .mcp.json with:
{
"mcpServers": {
"hypertool": {
"command": "npx",
"args": ["-y", "@toolprint/hypertool-mcp", "mcp", "run", "--mcp-config", ".mcp.hypertool.json"]
}
}
}
Step 3: Create Your First Toolset
Restart your AI and try:
You: "Create a toolset called 'coding' with git and docker tools"
AI: "Created 'coding' toolset with 15 focused tools"
You: "Switch to coding toolset"
AI: "Equipped! I now have just the tools needed for development"
That's it! Your AI is now focused and effective. ๐
๐ก Want automated setup? Try our interactive setup command - see Advanced Guide for details.
๐ Configuration Mode: HyperTool uses a smart Configuration Mode to keep toolset management separate from your operational tools. Learn more in the Configuration Mode Guide.
๐ญ Personas: Pre-configured Tool Bundles (NEW!)
Don't want to configure from scratch? Use personas - ready-to-use MCP server bundles with pre-built toolsets.
What are Personas?
Think of personas as "app bundles" for your AI - they come with:
- โ Pre-configured MCP servers
- โ Curated toolsets for specific workflows
- โ Everything you need to get started instantly
Quick Start with Personas
# 1. Clone the persona collection
git clone https://github.com/toolprint/awesome-mcp-personas
# 2. Add a persona (e.g., web-dev persona)
hypertool-mcp persona add awesome-mcp-personas/personas/web-dev
# 3. Run with the persona
npx -y @toolprint/hypertool-mcp mcp run --persona web-dev
That's it! No server configuration needed. The persona brings its own servers and toolsets.
๐ฆ Browse all available personas: awesome-mcp-personas
Available Personas
| Persona | Included Servers | Best For |
|---|---|---|
| web-dev | Git, Docker, Filesystem, Browser, Testing | Full-stack web development |
| data-scientist | Python, Jupyter, Database, Filesystem, Plotting | Data analysis & ML workflows |
| devops | Docker, Kubernetes, AWS, Terraform, Monitoring | Infrastructure & deployment |
| content-creator | Notion, Slack, Grammar, SEO, Social | Writing & content management |
| researcher | Perplexity, Arxiv, Wikipedia, Filesystem | Research & knowledge work |
Persona vs Standard Mode
# Standard Mode (use your existing MCP servers):
npx -y @toolprint/hypertool-mcp mcp run --mcp-config .mcp.hypertool.json
# Persona Mode (bundled servers + pre-built toolsets):
npx -y @toolprint/hypertool-mcp mcp run --persona web-dev
# Persona Mode with specific toolset:
npx -y @toolprint/hypertool-mcp mcp run --persona web-dev --equip-toolset frontend
๐ก Pro tip: Personas can be mixed with your existing servers! Add --mcp-config to include your custom servers alongside the persona's servers.
๐ Learn more: See the complete Personas Guide for detailed instructions, creating custom personas, and troubleshooting.
๐ Context Measurement (NEW!)
See exactly how much context each tool consumes. Optimize your toolsets with token estimates for every tool.
Active toolset showing token usage per tool
Why it matters:
- ๐ฏ Optimize context usage - Identify heavyweight tools consuming your context window
- ๐ Make informed decisions - See token costs before adding tools to toolsets
- ๐ Compare alternatives - Find lighter tools that do the same job
- ๐ก Budget your context - Understand exactly what you're exposing to your AI
How to use:
Ask your AI to use these MCP tools to see context information:
list-available-tools- Shows token estimates for all available toolsget-active-toolset- Shows token usage for your currently equipped toolset
Each tool displays estimated tokens and percentage of total context consumed. Perfect for building lean, efficient toolsets!
๐ฌ Demo
Hotswap toolsets across 100+ tools
Targeted toolsets across any number of MCPs. Swap to the best toolset for a goal with a tool call. Dynamic tool registration.
๐๏ธ How It Works
Before: Tool Chaos ๐ต
โโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
โ Claude/ โโโโถโ 50+ tools from 8 servers โ
โ Cursor โ โ โ Wrong picks โ
โ โ โ โ Slow decisions โ
โ โ โ โ Confused context โ
โโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
After: Expert Mode ๐ฏ
โโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโโโโโ
โ Claude/ โโโโถโ HyperTool โโโโถโ ALL Your Tools โ
โ Cursor โ โ (Local) โ โ (Same servers) โ
โโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโโโโโ
โ
โผ
โโโโโโโโโโโโโโโโโโโ
โ Smart Toolsets โ
โ ๐จ coding (5) โ โ "I'm coding now"
โ ๐ writing (3) โ โ "I'm writing now"
โ ๐ analysis (4) โ โ "I'm analyzing now"
โโโโโโโโโโโโโโโโโโโ
โ
Expert picks every time
What's a "Toolset"? Think Playlists for Your AI
Just like Spotify playlists organize your music, toolsets organize your AI tools:
ALL YOUR TOOLS (64 total) YOUR TOOLSETS
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโโโโโโ
โ ๐ณ Docker (19 tools) โ โ ๐จ "coding" โ
โ โข build_image โ โโโโโถ โ โข git.status โ
โ โข create_container โ โ โ โข git.commit โ
โ โข run_container โ โ โ โข docker.build โ
โ โข stop_container โ โ โ โข docker.run โ
โ โข [... 15 more] โ โ โ โข github.pr โ
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโค โ โโโโโโโโโโโโโโโโโโโโ
โ ๐ Git (12 tools) โโโโโค
โ โข status โ โ โโโโโโโโโโโโโโโโโโโโ
โ โข commit โ โ โ ๐ "writing" โ
โ โข push โ โโโโโถ โ โข notion.create โ
โ โข [... 9 more] โ โ โข slack.send โ
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโค โ โข grammarly.fix โ
โ ๐ Notion (8 tools) โโโโโโโ โโโโโโโโโโโโโโโโโโโโ
โ ๐ฌ Slack (6 tools) โ โ
โ ๐ Linear (10 tools) โ โ โโโโโโโโโโโโโโโโโโโโ
โ ๐งช Testing (9 tools) โ โโโถ โ ๐ "debugging" โ
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ โ โข logs.search โ
โ โข docker.logs โ
AI sees ALL 64 tools = confused ๐ต โ โข traces.view โ
โโโโโโโโโโโโโโโโโโโโ
AI sees 3-5 tools = focused ๐ฏ
๐ผ Real-World Toolsets
Create focused toolsets for different workflows:
๐จ Development Mode
"deep-coding": git + docker + filesystem (12 tools)
โ Everything you need for feature development
"code-review": git + github + linear (10 tools)
โ Review PRs, update tickets, merge with confidence
"debugging": logs + docker + traces + alerts (8 tools)
โ Find and fix issues fast
๐ Content Creation
"writing": notion + grammarly + slack (6 tools)
โ Blog posts, docs, and team updates
"research": perplexity + notion + filesystem (7 tools)
โ Deep dives with organized notes
๐ฌ Real Chat Example
You: "I need to debug our API"
AI: "I'll switch to the debugging toolset for better focus"
[Now has: logs, traces, curl, docker]
You: "Actually, let's write the incident report"
AI: "Switching to writing toolset"
[Now has: notion, slack, templates]
๐ก Pro tip: Start with 3-5 tools per toolset. Your AI will thank you!
๐ All Features
Explore everything HyperTool can do:
| Feature | Description | Guide |
|---|---|---|
| ๐ญ Personas | Pre-configured MCP server bundles with curated toolsets. Get started instantly with ready-to-use workflows for web-dev, data science, DevOps, and more. | Personas Guide |
| ๐ Server Groups | Organize MCP servers into logical groups. Launch related servers together, switch between projects, and maintain focused contexts. | Advanced Guide |
| ๐ Context Measurement | See token estimates for every tool. Optimize your toolsets by understanding exactly how much context each tool consumes. | Context Measurement |
| ๐ง Configuration Mode | Smart separation of toolset management from operational tools. Keep your AI focused on work, not configuration. | Configuration Mode Guide |
| ๐ฏ Dynamic Toolsets | Build, modify, and switch between toolsets on the fly. Adapt your AI's capabilities to match your current task. | Examples & Recipes |
| ๐ง Tool Annotations | Enhance tools with custom descriptions, examples, and context. Improve your AI's tool selection accuracy by 89%. | Advanced Guide |
| ๐ HTTP Mode | Run HyperTool as a long-lived HTTP server for persistent connections and faster response times. | Advanced Guide |
| ๐ Unlimited Servers | Connect as many MCP servers as you need. Break free from the 100-tool limit without sacrificing performance. | Quick Start |
โ FAQ
General Questions
Q: How is this different from just using MCP servers directly?
A: HyperTool lets you use unlimited MCP servers without hitting the 100-tool limit, and dynamically switches between focused toolsets for better AI performance.
Q: What's the difference between Personas and Standard Mode?
A: Standard Mode uses your existing MCP server configurations. Personas are pre-packaged bundles that include both MCP servers AND curated toolsets - perfect for getting started quickly or trying new workflows.
Q: Can I use multiple toolsets at once?
A: In stdio mode (default), use --equip-toolset <name> when launching. HTTP mode supports one active toolset but you can switch anytime.
Q: Where are my toolsets and configurations stored?
A: Everything is stored locally in ~/.toolprint/hypertool-mcp/:
- Personas:
~/.toolprint/hypertool-mcp/personas/ - Toolsets:
~/.toolprint/hypertool-mcp/toolsets/ - Preferences:
~/.toolprint/hypertool-mcp/config/
You can directly edit these files when HyperTool is not running.
Setup & Compatibility
Q: Does this work with Claude Desktop / Cursor / Claude Code?
A: Yes! Cursor has full hot-swapping support. Claude Desktop works with restart. Claude Code hot-swap coming soon.
Q: What if an MCP server goes down?
A: HyperTool monitors health and automatically reconnects when servers come back. Your toolsets stay intact.
Q: Can I share toolsets with my team?
A: Import/export is coming soon! For now, you can copy and share toolset files - they'll work if your team has the same MCP servers configured.
Q: How accurate are the token estimates in context measurement?
A: The estimates use BPE-based approximation for consistent relative comparisons between tools. They're perfect for understanding which tools consume more context, but not exact counts since different LLMs use different tokenizers.
Q: Does context measurement slow down my toolsets?
A: No! Token counts are cached and add less than 10ms overhead. You won't notice any performance impact.
Technical Questions
Q: How do I add tools from a new MCP server?
A: Just add the server to your .mcp.hypertool.json config. It's automatically available for toolsets.
Q: Can I use this in production?
A: Yes! For enterprise support, contact us.
๐ฎ App Compatibility
Works with ANY MCP-compatible app! HyperTool is a standard MCP server, so if your app supports MCP, it supports HyperTool.
Hot-swap Toolsets Without Restarts
| App | Status | How to Switch Toolsets |
|---|---|---|
| Cursor/VSCode | โ Full support | Switch toolsets instantly - no restart needed! |
| Claude Code | โณ Coming soon | Use --equip-toolset <name> flag (track progress) |
| Claude Desktop | โณ In progress | Restart app after switching toolsets |
๐ Learn More
- ๐ญ Personas Guide - Complete guide to using and creating personas
- ๐ฌ Research & Performance - Why focused toolsets work
- ๐ Advanced Features - Tool annotations, HTTP mode, CLI
- ๐ง Troubleshooting - Common issues and solutions
- ๐ Examples & Recipes - Toolset patterns for every workflow
๐ ๏ธ Development Setup
Prerequisites
- Node.js 18+
- Python 3.8+ (for pre-commit hooks)
Quick Setup
# Clone and install
git clone https://github.com/toolprint/hypertool-mcp.git
cd hypertool-mcp
just setup-dev # Installs dependencies and pre-commit hooks
Pre-commit Hooks
This project uses pre-commit hooks to ensure code quality:
# Install pre-commit hooks (included in setup-dev)
just setup-pre-commit
# Run hooks manually
just pre-commit-check # On staged files
just pre-commit-check-all # On all files
# Skip hooks for emergency commits (use sparingly)
SKIP=eslint,typescript git commit -m "emergency fix"
Available Commands
just build # Build the project
just test # Run tests
just lint # Run linting
just format # Format code
just typecheck # Check types
just pre-publish-checks # Run all quality checks
Service command
The hypertool-mcp service subcommand is currently disabled and will exit with a
notification when invoked.
๐ค Contributing
Found a bug? Have an idea? We'd love your help!
- ๐ Report issues
- ๐ก Share ideas
- ๐ง Submit PRs
๐ License
MIT License - see LICENSE file for details.
Built by developers who got tired of watching AI pick the wrong tools ๐ฏ
Built with โค๏ธ by Toolprint
ยฉ 2025 OneGrep, Inc.
FAQ
Common questions
Discussion
Questions & comments ยท 0
Sign In Sign in to leave a comment.