Skip to main content

Fast, accurate MCP tool retrieval engine for LLMs with semantic search

Project description

Agent-CoreX

🚀 Production-ready MCP tool retrieval engine + CLI for AI agents

Agent-CoreX solves a critical problem in LLM systems:

When you have hundreds of tools, how do you select the right few without blowing up the context window?


Agent-Corex on Product Hunt

Quick links: PyPI · Docs · Releases · GitHub


Install

Option 1 — Homebrew (macOS / Linux, no Python required)

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

Option 2 — Direct binary download (no Python required)

# macOS (arm64 — M1/M2/M3 and Intel via Rosetta 2)
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 — run PowerShell as Administrator:

# Installs to System32 — works from any terminal after this
Invoke-WebRequest `
  -Uri https://github.com/ankitpro/agent-corex/releases/latest/download/agent-corex-windows-x86_64.exe `
  -OutFile "$env:SystemRoot\System32\agent-corex.exe"
agent-corex --version

No admin access? See the per-user install →

Option 3 — pip (Python 3.8+)

pip install agent-corex

All three methods give you the same agent-corex CLI. Homebrew and the binary downloads require no Python installation.


Quick Start

# 1. Log in — opens your browser automatically, saves credentials locally
agent-corex login

# 2. Detect installed AI tools (Claude Desktop, Cursor, VS Code, etc.)
agent-corex detect

# 3. Inject agent-corex as an MCP server into all detected tools
agent-corex init

# 4. Verify everything is working
agent-corex status

# Run diagnostics if anything looks wrong
agent-corex doctor

Authentication

Agent-CoreX uses a browser-based login flow (like GitHub CLI or Vercel CLI) — no manual key copying required.

Login (recommended)

agent-corex login

What happens:

  1. A unique verification URL is generated and printed in the terminal
  2. Your browser opens automatically to https://agent-corex.com/cli-auth?code=...
  3. You log in with your Agent-CoreX account (email, Google, or GitHub)
  4. The CLI detects the completed login and saves your session locally
  5. Done — credentials are stored in ~/.agent-corex/config.json
$ agent-corex login

Logging in to Agent-CoreX...

Open this URL to complete login:

  https://agent-corex.com/cli-auth?code=a1b2c3d4-...

(Opened in your browser automatically)

Waiting for authentication.......

✓ Logged in successfully!
  User: you@example.com
  Session saved to ~/.agent-corex/config.json

You can now run:
  agent-corex sync
  agent-corex status

Login with API key (alternative)

If you prefer to use an API key directly:

# Paste your key (from the dashboard → API Keys)
agent-corex login --key acx_your_key_here

# Or: open the dashboard in browser, then prompt for key
agent-corex login --no-browser

Sync with dashboard

After logging in, sync your locally installed packs and servers with your Agent-CoreX dashboard:

agent-corex sync

What sync does:

  1. Pulls packs/servers you enabled in the dashboard → installs any missing locally
  2. Pushes your local install state → dashboard shows what's installed
agent-corex sync --push-only   # only push local state, skip pull/install

Logout

agent-corex logout

Clears all credentials (session tokens + API key) from ~/.agent-corex/config.json.

Local development

By default the CLI points to the production backend (https://www.agent-corex.com). To use a local backend instead:

agent-corex set-url http://localhost:8000

CLI Reference

Setup & Auth

Command Description
agent-corex login Log in via browser — opens browser, saves session automatically
agent-corex login --key <key> Log in with an API key directly
agent-corex login --no-browser Prompt for API key without opening browser
agent-corex logout Remove all stored credentials
agent-corex sync Sync installed packs/servers with dashboard
agent-corex sync --push-only Push local state to dashboard only
agent-corex keys Show masked API key + live backend verification

Tool Detection & Injection

Command Description
agent-corex detect Detect installed AI tools and show config paths
agent-corex init [--yes] Inject agent-corex MCP entry into all detected tools
agent-corex eject [--tool <t>] [--yes] Remove agent-corex from tool configs
agent-corex status Auth state, sync status, backend ping, injection status per tool

MCP Server Management

Command Description
agent-corex list List all MCP servers injected across detected tools
agent-corex registry Browse the installable MCP server catalog
agent-corex install-mcp <name> [--yes] Install any registry server into detected tools
agent-corex install-pack <name> [--yes] Install a curated pack of MCP servers
agent-corex update [--yes] Re-fetch registry and update injected server configs

Diagnostics

Command Description
agent-corex doctor Full health check: Python, PATH, config, backend, API key, injection
agent-corex health Quick backend health check
agent-corex version Print installed version

Gateway

Command Description
agent-corex serve Start the MCP gateway server (stdio — invoked by AI tools)

Tool Retrieval

Command Description
agent-corex retrieve "<query>" Semantic search for relevant tools

Supported AI Tools

Tool Platform
Claude Desktop macOS, Windows
Cursor macOS, Windows, Linux
VS Code macOS, Windows, Linux
VS Code Insiders macOS, Windows, Linux
VSCodium macOS, Windows, Linux

How It Works

Your AI tool (Claude / Cursor / VS Code)
         │  stdio
         ▼
  agent-corex serve        ← MCP gateway (single entry point)
         │
         ├─ retrieve_tools(query) → semantic search over all tools
         ├─ execute_tool(name, args) → routes call to correct MCP server
         └─ auth_check → validates Enterprise API key

The gateway exposes a small, fixed set of tools to Claude/Cursor instead of hundreds. This cuts context tokens by ~60% and keeps the LLM focused.


MCP Marketplace

Browse and install any MCP server with one command:

agent-corex registry              # browse catalog
agent-corex install-mcp github    # install GitHub MCP server
agent-corex install-mcp postgres  # install Postgres MCP server
agent-corex list                  # see all currently installed servers
agent-corex update                # pull latest configs from registry

Tool Retrieval (Python API)

from agent_core.retrieval.ranker import rank_tools

tools = rank_tools("edit a file", all_tools, top_k=5, method="hybrid")
# Returns: ["edit_file", "write_file", "create_file", ...]

Three ranking methods:

  • keyword — fast exact-match, no model required
  • embedding — semantic similarity via sentence-transformers
  • hybrid (recommended) — 30% keyword + 70% semantic

Releases & CI/CD

Every tag push triggers:

Workflow What it does
build-binaries.yml PyInstaller binaries for Linux x86_64, macOS arm64, Windows x86_64 + SHA256 sidecars
publish.yml Publishes to PyPI
update-homebrew-tap.yml Auto-patches the Homebrew formula with new version + SHA256s

Contributing

See CONTRIBUTING.md.

git clone https://github.com/ankitpro/agent-corex
cd agent-corex
pip install -e ".[dev]"
pytest

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

agent_corex-1.6.0.tar.gz (96.9 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

agent_corex-1.6.0-py3-none-any.whl (75.1 kB view details)

Uploaded Python 3

File details

Details for the file agent_corex-1.6.0.tar.gz.

File metadata

  • Download URL: agent_corex-1.6.0.tar.gz
  • Upload date:
  • Size: 96.9 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.1.0 CPython/3.13.7

File hashes

Hashes for agent_corex-1.6.0.tar.gz
Algorithm Hash digest
SHA256 e47f6223c47d699af97164feb421d6d37ad53150fbd225e48163186cbcf8feed
MD5 cec6ebe87bdb8802991d70010cfa68e2
BLAKE2b-256 d25bafa746d6bde2f348017cbec2ddc9672776594e0f960b6e26fcb8057c9122

See more details on using hashes here.

File details

Details for the file agent_corex-1.6.0-py3-none-any.whl.

File metadata

  • Download URL: agent_corex-1.6.0-py3-none-any.whl
  • Upload date:
  • Size: 75.1 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.1.0 CPython/3.13.7

File hashes

Hashes for agent_corex-1.6.0-py3-none-any.whl
Algorithm Hash digest
SHA256 0654ff4daffda402ddb7c8e46276bf51d917635da6fa0aaef690d268861170c4
MD5 7918c0c13bedccab7f1117d5b8cd1548
BLAKE2b-256 e3bb529f9706ab91a34ea97fc10acc273d1c6d923da7a8bc5cb769b308679bd5

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page