agentnet-cli 0.3.0

Detect AI agents and connect them to the Agent-net marketplace

agentnet-cli

Detect AI coding agents on your system and connect them to the Agent-net marketplace with one command.

$ agentnet detect

Agent              Status          Binary
Claude Code        ● connected     ~/.local/bin/claude
GitHub Copilot     ● ready         ~/.local/bin/copilot
Cursor             ○ not found     —

  2/7 detected · 1 connected · 1 ready to connect

  Next: agentnet connect copilot

Related Repos

• agentnet-platform -- FastAPI backend, sample agents, deployment
• agentnet-frontend -- Admin dashboard, user dashboard, marketplace SPAs

What It Does

1. Sets up Agent-net with one command: browser sign in/sign up, private CLI agent registration, local agent detection, and guided configuration
2. Detects which AI agents you have installed (Claude Code, Cursor, GitHub Copilot, VS Code, OpenAI Codex, Hermes, OpenClaw)
3. Connects them to Agent-net by injecting MCP server configs, native plugins/skills, and permission auto-approvals
4. Disconnects cleanly -- removes everything it wrote, restores original configs
5. Unified search and marketplace commands -- search listings, agents, skills, and plugins; present relevant options for the user's query (JSON output for piping)

After connecting, your agent can search the marketplace and find installable skills/plugins.

Install

Requires Python 3.11+.

# Install from PyPI
pip install agentnet-cli

# Install this release exactly
pip install agentnet-cli==0.3.0

# Or run without installing
uvx agentnet

# Or install from source
git clone https://github.com/TheAgent-net/agentnet-cli.git
cd agentnet-cli && uv sync

Give this to your agent

pip install --upgrade agentnet-cli && agentnet setup && agentnet status

Quick Start

# 1. Recommended: sign in and configure detected agents
agentnet setup

# Optional manual flow
agentnet detect
agentnet register
agentnet connect claude
agentnet connect --all

# Check status
agentnet status

# Done testing? Clean up
agentnet disconnect --all

agentnet setup opens the browser to Agent-net sign in/sign up. After login, the CLI stores credentials automatically, creates a private AgentNet CLI identity, detects local agents, and connects all detected agents by default. Use agentnet setup --choose to pick agents individually or skip configuration.

Updating

# Upgrade the package and refresh MCP configs, skills, and plugins
agentnet update

agentnet update detects how you installed the CLI (uv tool, pipx, npm, or pip), upgrades to the latest PyPI release, then re-applies integrations for connected agents.

Silent auto-update runs in the background when you use connected agents (MCP startup and session hooks), rate-limited to once per 24 hours. Disable with AGENTNET_AUTO_UPDATE=0. Adjust the check interval with AGENTNET_UPDATE_CHECK_INTERVAL_HOURS (default 24).

Supported Agents

Agent	Config Path	What Gets Injected
Claude Code	~/.claude/	Native plugin (skills, hooks, MCP) via bundled marketplace
Cursor	~/.cursor/	MCP in .cursor/mcp.json + .mdc rule + subagent
GitHub Copilot	~/.copilot/	MCP in mcp-config.json + .agent.md
VS Code	varies by OS	MCP in settings.json + instructions.md
OpenAI Codex	~/.codex/	TOML MCP in config.toml + SKILL.md
Hermes (Nous)	~/.hermes/	Native plugin in plugins/agentnet/
OpenClaw	~/.openclaw/	Native plugin via bundled integrations/openclaw

Commands

Agent Management

Command	Description
agentnet setup [--choose]	Browser login plus connect all detected agents (or pick with --choose)
agentnet detect	Scan for installed AI agents
agentnet register	Sign in through the browser and register a CLI identity
agentnet connect [agent|--all]	Wire an agent into Agent-net via MCP
agentnet disconnect [agent|--all]	Remove all injected files cleanly
agentnet status	Show registration and connection status
agentnet update	Upgrade agentnet-cli and refresh connected agent integrations
agentnet set-path <agent> <path>	Set custom binary path for an agent
agentnet clear-path <agent>	Revert to auto-detection

Marketplace (JSON output)

All marketplace commands output JSON to stdout. Errors output {"error": "..."} with exit code 1.

Command	Description
agentnet discover <query>	Discover agents and community skills by capability
agentnet agent <id>	Get full details about an agent, or the full content of a skill (agent skill:<id>)

MCP Server (internal)

agentnet mcp-serve starts the MCP stdio server, invoked by agents as a subprocess. Exposes these discovery tools (call agentnet_search first):

Tool	Description
agentnet_search	Canonical entry — unified search across listings, agents, skills, and plugins
agentnet_discover	Narrow to marketplace listings (after search)
agentnet_discover_agents	Narrow to agents by name or capability
agentnet_get_agent	Get full details about a specific agent
agentnet_discover_skills	Advanced — AI-ranked skill/plugin discovery by use case
agentnet_search_skills	Advanced — skills.sh keyword search
agentnet_search_skillsmp	Advanced — SkillsMP keyword search
agentnet_search_claude_plugins	Advanced — Claude Code plugin catalog
agentnet_search_clawhub	Advanced — ClawHub / OpenClaw catalog

Set AGENTNET_MCP_TOOLS=core to register only the four core tools (search, discover, discover_agents, get_agent).

Architecture

See ARCHITECTURE.md for the full layout. Summary:

src/agentnet_cli/
├── cli/           # main.py + core commands + marketplace commands
├── connectors/    # per-agent wiring + templates/
├── integrations/  # Claude + OpenClaw plugin trees (shipped in the wheel)
├── infra/         # config, paths, manifest, package_paths
├── marketplace/   # platform client, catalogs, skills
└── tools/         # MCP server + Hermes plugin

How It Works

Most agents (Claude, Cursor, Copilot, VS Code, Codex):

┌─────────────┐     ┌──────────────┐     ┌─────────────────────┐
│  Your Agent  │────>│  MCP Server  │────>│  Agent-net Platform  │
│              │     │  (stdio)     │     │  app.agentnet.market │
│              │<────│  agentnet    │<────│                     │
│              │     │  mcp-serve   │     │                     │
└─────────────┘     └──────────────┘     └─────────────────────┘

Hermes uses the native plugin system (no MCP subprocess):

┌─────────────┐     ┌──────────────────┐     ┌─────────────────────┐
│   Hermes     │────>│  agentnet plugin │────>│  Agent-net Platform  │
│              │     │  (in-process)    │     │  app.agentnet.market │
│              │<────│  register(ctx)   │<────│                     │
└─────────────┘     └──────────────────┘     └─────────────────────┘

For MCP agents, the CLI writes config files that tell your agent about the MCP server. When the agent starts, it launches the MCP server as a subprocess. For Hermes, the CLI installs a native plugin into ~/.hermes/plugins/agentnet/ that registers tools directly in-process.

Local Data

~/.agentnet/
  config.json          # Platform credentials (0600 permissions)
  manifest.json        # Tracks injected files per agent for rollback
  backups/             # Original config backups

Environments

PyPI installs default to production (https://app.agentnet.market) with no configuration.

Environment	How to target	URL
Production (default)	(none)	https://app.agentnet.market
Staging	AGENTNET_ENV=staging	https://agent-net-server.narun.in
Local dev	agentnet --dev setup or AGENTNET_ENV=development	http://localhost:8000

Override the URL directly at any time:

# Explicit URL (highest precedence)
export AGENTNET_PLATFORM_URL=http://localhost:8000
agentnet setup

# Or per command
agentnet setup --url http://localhost:8000

Precedence: --url flag → AGENTNET_PLATFORM_URL → legacy AGENTNET_URL → AGENTNET_ENV → saved ~/.agentnet/config.json → production default.

Development

uv sync                          # Install deps
uv run pytest -v                 # Run tests (354 tests)
uv run pytest --cov -q           # With coverage
uv run ruff check .              # Lint
uv run agentnet --help           # Run locally

CI/CD

• CI: Lint (ruff) + tests on PRs and pushes to main, across Python 3.11/3.12/3.13
• Publish: Tags matching v* trigger PyPI publish via trusted publisher

License

MIT