πŸš€ Agent-Corex

Intelligent Tool Selection for LLMs

Agent-Corex - Intelligent Tool Selection for LLMs

Fast, accurate MCP tool retrieval engine with hybrid semantic + keyword ranking

Agent-Corex

Intelligent Tool Selection for LLM-Based Systems

Stop feeding your LLM tools it doesn’t need. Agent-Corex intelligently selects the top N most relevant tools in milliseconds using hybrid semantic + keyword ranking.


Install

Homebrew (macOS / Linux β€” no Python required)

brew tap ankitpro/agent-corex
brew install agent-corex

Direct binary (no Python required)

Platform Command
macOS arm64 (M1/M2/M3 + Intel via Rosetta) curl -fsSL https://github.com/ankitpro/agent-corex/releases/latest/download/agent-corex-macos-arm64 -o /usr/local/bin/agent-corex && chmod +x /usr/local/bin/agent-corex
Linux x86_64 curl -fsSL https://github.com/ankitpro/agent-corex/releases/latest/download/agent-corex-linux-x86_64 -o /usr/local/bin/agent-corex && chmod +x /usr/local/bin/agent-corex
Windows x86_64 Download .exe

pip (Python 3.8+)

pip install agent-corex

Full installation guide β†’


The Problem We Solve

When building LLM agents with access to hundreds of tools, including all of them in the system prompt causes:

  • Context Bloat β€” 30K+ tokens per request
  • Higher Costs β€” 68% more expensive per query
  • Slower Inference β€” 4–5x latency increase
  • Model Confusion β€” LLM picks random tools when overwhelmed

Agent-Corex solves this by intelligently selecting only the 5–10 tools your LLM actually needs.


Quick Stats

Metric Impact
Token Reduction 50–75% fewer tokens
Speed Improvement 3–5x faster inference
Cost Savings 68% reduction per query
Selection Accuracy 92% precision (top-5)

One-Command Setup

Connect Agent-CoreX to Claude Desktop, Cursor, VS Code, or VS Code Insiders:

agent-corex login --key acx_your_key
agent-corex init

Agent-CoreX detects your installed tools, backs up their configs, and safely merges the gateway entry β€” existing MCP servers are never touched.

Full setup guide β†’


Key Features

Three Ranking Methods

Method Speed Best For
Keyword <1ms Instant filtering
Hybrid 50–100ms Recommended
Embedding 50–100ms Maximum semantic accuracy

MCP Marketplace

agent-corex registry              # browse catalog
agent-corex install-mcp github    # install GitHub MCP server
agent-corex list                  # see installed servers
agent-corex update                # update all server configs

Multiple Interfaces

Python SDK  β†’  REST API  β†’  CLI Tool  β†’  MCP Gateway

Documentation


Community & Support


License

Agent-Corex is released under the MIT License. Author: Ankit Agarwal