agentx

Install, chain, and orchestrate AI agents from the terminal.

---

agentx lets you discover, install, chain, and orchestrate AI agents from the terminal. Pipe one agent's output into another with standard Unix pipes — research, analyze, write, and ship in a single command. Agents are reusable packages for Claude Code that bundle system prompts, MCP servers, and secrets.

Features

• Chain agents with pipes - agentx run agent-a --quiet | agentx run agent-b "use this"
• Run agents - Execute agents locally with agentx run <agent> "prompt"
• Install from registry - One command install: agentx install @scope/agent
• Schedule agents - Cron-based scheduling: agentx schedule start <agent>
• Search & discover - Find agents via CLI or browse agentx.dev
• Publish agents - Share your agents with agentx publish
• Scaffold agents - Create new agents with agentx init
• Encrypted secrets - AES-256-GCM encrypted secrets per agent
• MCP integration - Agents declare MCP servers for tool access

Prerequisites

• Node.js >= 18
• Claude CLI installed and authenticated (npm install -g @anthropic-ai/claude-code)

Install

npm install -g @knid/agentx

Verify your setup:

agentx doctor

Quick Start

# Search for an agent
agentx search "data analysis"

# Install it
agentx install @agentx/data-analyst

# Run it
agentx run data-analyst "analyze trends in this data" --file sales.csv

# Chain agents — research, then write
agentx run web-researcher --quiet "2026 AI trends" \
  | agentx run writing-assistant "turn this into a blog post"

# Interactive mode
agentx run data-analyst -i

Commands

Command	Description
agentx run <agent> [prompt]	Run an agent with a prompt
agentx install <agent>	Install an agent from the registry
agentx uninstall <agent>	Remove an installed agent
agentx update [agent|--all]	Update agents to latest versions
agentx list	List installed agents
agentx search <query>	Search the agent registry
agentx trending	Show trending agents
agentx info <agent>	Show agent details
agentx init	Scaffold a new agent project
agentx validate	Validate an agent manifest
agentx test	Test an agent locally
agentx publish	Publish an agent to the registry
agentx schedule start <agent>	Start an agent's cron schedule
agentx schedule stop <agent>	Stop an agent's schedule
agentx schedule list	List all active schedules
agentx schedule logs <agent>	View execution logs for a scheduled agent
agentx schedule resume	Resume all schedules after restart
agentx configure <agent>	Configure secrets for an agent
agentx login	Authenticate with GitHub
agentx logout	Clear authentication
agentx whoami	Show current user
agentx doctor	Check system requirements
agentx config	Manage global configuration

Create Your Own Agent

# Scaffold a new agent
agentx init

# Edit the generated files
# - agent.yaml    (manifest: name, version, MCP servers, permissions)
# - system-prompt.md  (the system prompt for Claude)

# Validate it
agentx validate

# Test locally
agentx run . "test prompt"

# Publish to the registry
agentx login
agentx publish

Agent Manifest (agent.yaml)

name: my-agent
version: 1.0.0
description: A helpful agent
author: "@yourusername"
category: productivity
permissions:
  filesystem: true
  network: true
pre_run:
  - command: my-bridge
    background: true
mcp_servers:
  filesystem:
    command: npx
    args: ["-y", "@modelcontextprotocol/server-filesystem", "./"]
secrets:
  - name: API_KEY
    description: API key for the service
    required: true
schedule:
  - name: "Daily report"
    cron: "0 9 * * 1-5"
    prompt: "Generate the daily report"

Agent Chaining

Agents write to stdout, so you can chain them with standard Unix pipes. The output of one agent becomes the input context for the next — build multi-step AI workflows in a single line:

# Research a topic, then create a Notion page from the results
agentx run web-researcher --quiet "2026 AI trends" \
  | agentx run notion-agent "create a new page summarizing this research"

# Scan for vulnerabilities, then create a Linear issue for each finding
agentx run security-scanner --quiet "audit src/auth/ for vulnerabilities" \
  | agentx run linear-agent "create a bug for each critical finding"

# Analyze data, then draft a report
agentx run data-analyst --quiet "summarize quarterly revenue" --file q4.csv \
  | agentx run writing-assistant "turn this into an executive summary"

# Review code, then send the review to Slack
agentx run code-reviewer --quiet "review the latest changes in src/api/" \
  | agentx run slack-agent "post this code review summary to #engineering"

# Three-step pipeline: research → rewrite → post
agentx run web-researcher --quiet "latest React best practices 2026" \
  | agentx run writing-assistant --quiet "rewrite as a concise team guide" \
  | agentx run slack-agent "post this to #frontend"

Use --quiet on intermediate agents to suppress headers/footers and pipe only the raw output. The last agent in the chain can run without --quiet to display formatted output.

Scheduling

Agents can declare cron-based schedules in agent.yaml. A shared background daemon runs on your machine and executes agents at the specified times.

# Start an agent's schedule
agentx schedule start slack-agent

# View active schedules
agentx schedule list

# Check execution logs
agentx schedule logs slack-agent

# View all past runs
agentx schedule logs slack-agent --all

# Stop a schedule
agentx schedule stop slack-agent

# Resume all schedules after a restart
agentx schedule resume

The daemon automatically retries failed runs (up to 2 retries with backoff), rotates logs (keeps last 50 per agent), and cleans up when all schedules are stopped.

Official Starter Agents

14 agents across all 10 categories — install any with agentx install @agentx/<name>:

Agent	Category	Description
@agentx/gmail-agent	communication	Email assistant with Gmail MCP
@agentx/slack-agent	communication	Messaging assistant with Slack MCP
@agentx/whatsapp-agent	communication	WhatsApp messaging via local bridge
@agentx/github-agent	devtools	PR and issue management with GitHub MCP
@agentx/code-reviewer	devtools	Code review with GitHub + filesystem MCP
@agentx/data-analyst	data	CSV/JSON data analysis with filesystem MCP
@agentx/postgres-agent	data	PostgreSQL query and schema explorer
@agentx/web-researcher	research	Web search and synthesis with Brave + Fetch
@agentx/notion-agent	productivity	Notion workspace and database management
@agentx/linear-agent	productivity	Linear issue tracking and sprint management
@agentx/sentry-agent	monitoring	Error triage and stack trace analysis
@agentx/puppeteer-agent	automation	Browser automation, screenshots, and scraping
@agentx/writing-assistant	writing	Proofreading, drafting, and document editing
@agentx/security-scanner	security	Vulnerability scanning and dependency auditing

Configuration

Global config is stored at ~/.agentx/config.yaml:

# View all config
agentx config list

# Get a value
agentx config get registry

# Set a value
agentx config set telemetry false

Key	Default	Description
registry	https://registry.agentx.dev	Registry URL
claude_path	claude	Path to Claude CLI
default_output	text	Default output format (text/json)
telemetry	true	Enable anonymous telemetry
auto_update	true	Auto-check for updates

Project Structure

packages/
  cli/       # agentx CLI (npm package)
  web/       # agentx.dev website and registry API
  agents/    # Official starter agents

Development

# Clone the repo
git clone https://github.com/agentx-dev/agentx.git
cd agentx

# Install dependencies
npm install

# Build the CLI
npm run build --workspace=packages/cli

# Run tests
npm test --workspace=packages/cli

# Type check
npx tsc --noEmit --project packages/cli/tsconfig.json

# Link for local development
cd packages/cli && npm link

See CONTRIBUTING.md for full development guidelines.

License

MIT