cablate/Agentic-MCP-Skill

View on GitHub ↗

Agentic-MCP, Progressive MCP client with three-layer lazy loading. Validates AgentSkills.io pattern for efficient token usage. Use MCP without pre-install & wasting full-loading

39 ★4 forksTypeScriptUpdated 8mo ago

What you need to know

Experimental concept validation porting the AgentSkills.io progressive-disclosure pattern to MCP. A socket-based daemon manages MCP servers with three-layer loading (metadata, tool list, individual tool schemas) to reduce token usage. npm package @cablate/agentic-mcp with CLI.

Install

npm install -g @cablate/agentic-mcp
agentic-mcp daemon start

Usage

  • •agentic-mcp metadata playwright (Layer 1: server status)
  • •agentic-mcp list playwright (Layer 2: tool list)
  • •agentic-mcp schema playwright browser_navigate (Layer 3: tool format)
  • •agentic-mcp call playwright browser_navigate --params '{"url": "https://example.com"}'
  • •agentic-mcp daemon reload after editing mcp-servers.json

Key features

  • ✓Three-layer progressive disclosure (metadata/list/schema) with ~86% token savings example
  • ✓Long-running daemon with hot reload and shared sessions over socket (NDJSON)
  • ✓CLI to metadata, list, schema, call tools
  • ✓Config via mcp-servers.json

Best for

Validating token-efficient MCP usage patterns via progressive disclosure

Caveats

  • ⚠Very early AI-assisted demo, not recommended for production use
  • ⚠Test coverage focuses on core modules; daemon integration tests deferred
Clients: CLI · AI agents
Documentation ↗

Reviewed 2026-08-11

Topics

agentskillagentskillslazy-loadingmcpmcp-servermodel-context-protocolskillskills
Stars
39★
Forks
4
Language
TypeScript
License
NOASSERTION
Created
2026-01-12
Last push
2026-01-21