Agentic-MCP-Skill
by cablate·★ 32·Score 40
A progressive MCP client with three-layer lazy loading that reduces token usage by up to 86%.
Overview
Agentic-MCP-Skill is an MCP client that implements a three-layer progressive disclosure pattern to validate efficient token usage. Instead of loading all MCP server tools at once, it allows users to load only basic server information first, then specific tool lists, and finally complete tool schemas as needed. This approach can significantly reduce context token usage by up to 86% when compared to traditional MCP usage methods. The project features a daemon architecture that maintains persistent MCP connections and supports hot reloading of configurations without restarting.
Try asking AI
After installing, here are 5 things you can ask your AI assistant:
When to choose this
Choose this MCP client when working with MCP servers that have many tools but you only need to use occasionally, or when token usage efficiency is a critical concern.
When NOT to choose this
Don't choose this for production environments as it's explicitly labeled as an early concept validation with known issues. Consider mature MCP implementations instead.
Tools this server exposes
6 tools extracted from the READMEbrowser_navigateNavigate to a website using browser automation
browser_take_screenshotTake a screenshot of the current browser page
browser_clickClick on a specific element in the browser
metadataCheck MCP server status and basic information
listList available tools on an MCP server
schemaView the complete input format for a specific tool
Comparable tools
Installation
Installation
# Install globally
npm install -g @cablate/agentic-mcp
# Verify installation
agentic-mcp --version
# Start daemon
agentic-mcp daemon startClaude Desktop Configuration
Add to Claude's claude_desktop_config.json:
{
"mcpServers": {
"agentic-mcp": {
"command": "agentic-mcp",
"args": ["daemon", "start"]
}
}
}FAQ
- How does the three-layer loading reduce token usage?
- Instead of loading all MCP tools at once (~6000 tokens), the three-layer approach loads only server metadata (~50 tokens), tool lists (~200 tokens), and specific tools (~300-500 tokens each) as needed, potentially reducing usage by up to 86%.
- What MCP servers are supported?
- Any MCP server compatible with stdio transport is supported. Examples include playwright, filesystem, github, and custom servers as configured in the mcp-servers.json file.
Compare Agentic-MCP-Skill with
Last updated · Auto-generated from public README + GitHub signals.