Most web browsing recipes in the cookbooks use Playwright/Chrome to fetch page content before sending it to Claude. This works but sends 50K+ tokens of HTML noise per page.
Plasmate is an open-source browser engine that outputs semantic content, compressing pages by 10-16x. It also ships as an MCP server.
A cookbook recipe could show:
- Install Plasmate:
pip install plasmate
- Fetch a page:
plasmate fetch <url> -> clean JSON
- Send the SOM to Claude (16x fewer tokens, 94% cost savings)
- Compare the same task with raw HTML vs SOM
We have Jupyter notebooks ready if useful: https://github.com/plasmate-labs/notebooks
Happy to contribute the recipe if there's interest. Apache 2.0.
Most web browsing recipes in the cookbooks use Playwright/Chrome to fetch page content before sending it to Claude. This works but sends 50K+ tokens of HTML noise per page.
Plasmate is an open-source browser engine that outputs semantic content, compressing pages by 10-16x. It also ships as an MCP server.
A cookbook recipe could show:
pip install plasmateplasmate fetch <url>-> clean JSONWe have Jupyter notebooks ready if useful: https://github.com/plasmate-labs/notebooks
Happy to contribute the recipe if there's interest. Apache 2.0.