OpenClaw Integration

OpenClaw integration for Plasmate — install as a skill to give any OpenClaw agent a fast, token-efficient browser engine.

Skill repo: plasmate-labs/skill-openclaw

Installation

1. Install Plasmate

curl -fsSL https://plasmate.app/install.sh | sh

2. Install the skill

clawhub install plasmate

Or manually copy integrations/openclaw/SKILL.md to ~/.openclaw/skills/plasmate/SKILL.md.

3. Install the `pf` wrapper

cp integrations/openclaw/scripts/pf /usr/local/bin/pf
chmod +x /usr/local/bin/pf

Quick Start

Replace web_fetch calls with pf:

# Before
web_fetch https://docs.stripe.com/api

# After — ~96% fewer tokens, stats logged automatically
pf https://docs.stripe.com/api

pf wraps plasmate fetch, prints timing + token savings to stderr, and appends a stat entry to ~/.plasmate/fetch-stats.jsonl.

Token Savings

Real-world benchmark (SOM vs raw HTML, 12 sites):

Site	Plasmate	Raw HTML	Savings
Vercel docs	2,206 tok	556,464 tok	99.6%
Stripe API	12,699 tok	301,604 tok	95.8%
Next.js docs	15,350 tok	198,307 tok	92.3%
Stack Overflow	41,699 tok	289,090 tok	85.6%
Wikipedia	25,448 tok	147,538 tok	82.8%

1.56M tokens saved across 10 test fetches. Plasmate is most effective on SPAs and content-heavy pages.

MCP Integration

For multi-step browsing, run Plasmate as an MCP server:

plasmate mcp

Add to your agent's MCP config:

{
  "servers": {
    "plasmate": {
      "command": "plasmate",
      "args": ["mcp"],
      "transport": "stdio"
    }
  }
}

Available MCP tools: fetch_page, extract_text, screenshot_page, open_page, navigate_to, click, type_text, select_option, scroll, toggle, clear, evaluate, close_page.

CDP Mode (Puppeteer-compatible)

Run Plasmate as a CDP server to replace Chrome in existing Puppeteer/Playwright workflows:

plasmate serve --protocol cdp --port 9222
export BROWSER_WS_ENDPOINT="ws://127.0.0.1:9222"

Viewing Fetch Stats

The pf wrapper logs every fetch:

python3 - << 'EOF'
import json, os
log = os.path.expanduser("~/.plasmate/fetch-stats.jsonl")
entries = [json.loads(l) for l in open(log) if l.strip()]
n = len(entries)
saved = sum(e.get("tokens_saved_est", 0) for e in entries)
print(f"{n} fetches | {saved:,} tokens saved")
EOF