Browse the web from your OpenClaw agent via Plasmate.
Fast. Structured. Token-efficient.
An OpenClaw skill that gives agents web browsing capabilities via Plasmate's Semantic Object Model. Instead of raw HTML (tens of thousands of tokens), your agent gets structured JSON that's 10-800x smaller.
clawhub install plasmateOr clone this repo into your OpenClaw skills directory.
plasmate/
├── SKILL.md # Skill definition and usage guide
├── scripts/
│ └── awp-browse.py # AWP client for agent web interactions
└── references/
└── auth-flow.md # Authenticated browsing documentation
Install Plasmate:
curl -fsSL https://plasmate.app/install.sh | shOnce the skill is installed, your OpenClaw agent can:
Navigate and extract structured content:
python3 scripts/awp-browse.py navigate "https://news.ycombinator.com"Click interactive elements:
python3 scripts/awp-browse.py click "https://example.com" --ref "e12"Type into fields:
python3 scripts/awp-browse.py type "https://example.com" --ref "e5" --text "search query"Extract structured data (JSON-LD, OpenGraph):
python3 scripts/awp-browse.py extract "https://example.com"The script auto-starts a Plasmate server if one isn't running.
For sites requiring login (X, GitHub, enterprise SaaS):
- Log in via your browser
- Export cookies with the Plasmate extension or grab them from dev tools
- Store them:
plasmate auth set x.com --ct0 <val> --auth-token <val> - Agent browses with
--profile x.com
See references/auth-flow.md for details.
| Plasmate | Chrome | |
|---|---|---|
| Per page | 4-5 ms | 252 ms |
| Memory (100 pages) | ~30 MB | ~20 GB |
| Output tokens | SOM (10-800x smaller) | Raw HTML |
| Binary size | 43 MB | 300-500 MB |
Your agent reasons better with less noise, and your token budget goes further.
- Plasmate - The browser engine for agents
- Plasmate Extension - Chrome extension for cookie export
- OpenClaw - The agent platform
MPL-2.0 - Modifications to these files must stay open. Use freely in any project.