Plasmate Documentation - Plasmate Docs | The Better Browser for AI Agents

# Plasmate > The browser engine for AI agents. HTML in, Semantic Object Model out. Plasmate is an open source headless browser engine written in Rust, purpose-built for AI agents. Instead of rendering pixels, it compiles web pages into a Semantic Object Model (SOM): structured JSON with typed regions, interactive elements with stable IDs, and clean text content. 17.5x average token compression across 51 real websites. ## Key facts - Latest: v0.5.0 (March 2026) - 25K lines Rust, 224+ tests, Apache-2.0 licensed - 17.5x average token compression (benchmarked across 51 real URLs) - 94% token savings on content-heavy sites (Stripe docs 95.8%, Vercel docs 99.6%) - 4ms compile time per page, 200ms with daemon mode warm - 30MB memory for 100 pages (vs ~20GB for headless Chrome) - First browser/web tool on the official MCP Registry - 13 MCP tools: fetch_page, extract_text, extract_links, open_page, navigate_to, click, type_text, select_option, scroll, toggle, clear, evaluate, close_page - Also speaks CDP (Puppeteer/Playwright compatible) and AWP (native agent protocol) - SDKs: Python (PyPI), Node.js (npm), Go, Rust (crates.io), WASM (npm) - Integrations: LangChain, LlamaIndex, CrewAI, Browser Use, AutoGen, Smolagents, Pi/oh-my-pi, n8n, OpenClaw, Scrapy ## Install (30 seconds) Claude Code (one command, no config): ``` claude mcp add plasmate -- npx plasmate-mcp ``` OpenClaw (one command): ``` clawhub install plasmate ``` Cursor / Claude Desktop / any MCP client: ```json { "mcpServers": { "plasmate": { "command": "npx", "args": ["plasmate-mcp"] } } } ``` Binary install (for CLI usage, not needed for MCP): ``` curl -fsSL https://plasmate.app/install.sh | sh ``` Or: `cargo install plasmate` | `npm install -g plasmate` | `pip install plasmate` ## CLI usage ```bash plasmate fetch https://example.com # Full SOM JSON plasmate fetch https://example.com --format text # Plain text plasmate fetch https://example.com --format ma