@zagents/host-openclaw
v0.1.3
Published
OpenClaw (personal AI assistant) host adapter for zagents-generator (MCP via ~/.openclaw/openclaw.json + workspace skills)
Maintainers
Readme
@zagents/host-openclaw
OpenClaw host adapter for the zagents-generator project.
OpenClaw — Personal AI Assistant. Any OS. Any Platform. The lobster way. 🦞
Generates the per-gemini files an OpenClaw-targeted gemini needs:
openclaw.json— MCP server snippet to merge into~/.openclaw/openclaw.jsonSKILL.md— workspace skill (placed at~/.openclaw/workspace/skills/<name>/SKILL.md)install-openclaw.sh— runbook script (idempotent)
Usage
import adapter from '@zagents/host-openclaw';
const config = adapter.generateConfig({
name: 'my-bot',
description: 'My personal lobster assistant',
systemPrompt: 'You are a helpful agent.',
mcpServers: [{ name: 'my-bot', command: ['npx', '-y', 'my-bot', 'mcp'] }],
});
// config['openclaw.json'] = '<JSON>'
// config['SKILL.md'] = '<YAML+markdown>'
// config['install-openclaw.sh'] = '<bash>'OpenClaw integration surface
| Aspect | Value |
|---|---|
| Install | npm install -g openclaw@latest then openclaw onboard --install-daemon |
| Main config | ~/.openclaw/openclaw.json (JSON, not TOML/YAML) |
| Skills | ~/.openclaw/workspace/skills/<name>/SKILL.md (YAML frontmatter + markdown) |
| First-class tools | browser, canvas, nodes, cron, sessions |
| MCP servers | register under mcp_servers in openclaw.json |
| Node version | ≥ 22.19 / 24 |
| License | MIT |
What's different from the other host adapters
| | OpenClaw | Claude Code | Codex | Hermes | pi.dev | |---|---|---|---|---|---| | Config format | JSON | JSON | TOML | YAML | TypeScript module | | Hooks system | (none documented) | rich, 5-handler | none | none | none | | MCP support | yes | yes | yes | yes | no (by design) | | Multi-platform messaging | yes (built-in) | no | no | no | no |
OpenClaw is the only host with built-in multi-platform messaging (WhatsApp/Telegram/Slack/Discord). For a gemini that needs to push notifications to the user across channels, OpenClaw is the natural fit.
License
MIT
