Show HN: PageMap – MCP server that compresses web pages from 100K to 2-5K tokens (github.com)

PageMap, a new server for compressing web pages, was recently announced, enabling the conversion of approximately 100K-token HTML documents into a structured map of just 2-5K tokens. This innovative technology significantly enhances AI agents' ability to interact with web pages, as it allows them to comprehend and execute actions on the content with 97% fewer tokens consumed. Unlike existing tools, PageMap provides an interactive experience by enabling agents to click, type, or select elements on the page, thereby overcoming the limitations faced by tools like Firecrawl and Jina that only offer read-only access. The introduction of PageMap is particularly significant for the AI/ML community, as it paves the way for more efficient multi-page sessions without exceeding context window limits. Benchmarked across various e-commerce sites, PageMap achieved a remarkable task success rate of 95.2% while incurring lower costs than traditional methods. It employs sophisticated mechanisms for processing and filtering HTML, maintaining security against potential input vulnerabilities through robust defenses. With features like automatic locale detection and built-in support for internationalization, PageMap represents a major advancement in making web interactions manageable and actionable for AI agents, ultimately driving the evolution of web intelligence in AI applications.