Compress.new

Compress.new is a hosted MCP server built for one specific, high-value job: turning messy, ad-laden webpages into clean, LLM-ready markdown before that content ever reaches your AI agent's context window. Instead of dumping raw HTML — with its navigation bars, scripts, tracking pixels, and boilerplate — into a prompt, agents call Compress.new to fetch a URL and get back a trimmed, well-structured markdown document that preserves headings, lists, tables, and links while stripping everything that inflates token counts. For teams running research agents, RAG pipelines, or automated content workflows, this translates directly into lower token costs and faster, more reliable downstream reasoning, since the model spends its context budget on substance instead of noise.

Within the BusinessMCP.com model, Compress.new plugs into your unified MCP server alongside your other business tools — CRM, ad platforms, databases, and revenue data — so any AI agent (Claude, GPT, Gemini, or a custom model) can call it through the same /api/mcp endpoint with a single Bearer mcph_* key. There's no separate scraping library to install, no headless browser to maintain, and no per-tool auth to juggle. Connect it once in your BusinessMCP dashboard and it becomes just another callable capability your agents can reach for whenever a task requires reading live web content — whether that's pulling a competitor's pricing page, ingesting a documentation site for a support bot, or feeding a news article into a summarization workflow.

Because the server is model-agnostic and cookieless by design, it fits naturally into GDPR-conscious stacks and doesn't tie you to a single AI vendor. Swap models, run multi-agent setups, or A/B test different LLMs against the same webpage-to-markdown pipeline without touching your integration. This matters for teams that treat token spend as a real line-item cost: converting a webpage to markdown before it hits the model is one of the simplest, most repeatable ways to reduce LLM token costs across research, monitoring, and content-generation agents that otherwise re-fetch and re-process the same bloated pages repeatedly.

Typical adopters include product and market-research teams building agents that scan multiple sources per query, support and knowledge-base bots that need clean reference material from external docs, and finance or operations teams (hence its home in the Finance category) using AI agents to monitor filings, investor pages, or vendor terms without paying markdown-cleanup token overhead on every run. Because BusinessMCP surfaces usage across all connected tools in one business-intelligence dashboard, you can see exactly how often Compress.new is invoked relative to your other MCP tools and correlate that with downstream agent performance and cost.

Compress.new is best paired with fetch- and browser-style MCP tools for retrieval, filesystem or memory tools for persisting the cleaned output, and structured-data tools for anything that needs to go beyond plain reading. Used together inside one hosted MCP server, they let an agent fetch, clean, store, and act on web content in a single coherent workflow rather than stitching together disparate scraping and parsing services.