AgentReady MCP - The MCP that puts Claude, Cursor & Windsurf on a token diet
by•
New: AgentReady now works inside your AI assistant. One MCP server (npx agentready-cloud-mcp) gives Claude Code, Cursor and Windsurf three tools: compress long context before it is read, open any URL as clean Markdown instead of raw HTML, and shrink JSON into TOON. Building an app? The new self-hosted gateway squeezes every OpenAI and Anthropic call while your API key never leaves your machine. Free during beta.


Replies
Hey Product Hunt! 👋 Christian here.
In February you took AgentReady to #9 of the day. Since then you've pushed 2.1M+ tokens through it across 10,000+ API calls, and 1.2M of those tokens never hit an LLM bill. Thank you 🙏
Today AgentReady moves to where you actually work: inside your AI assistant.
🧩 AgentReady MCP (one command, free):
claude mcp add agentready -e AGENTREADY_API_KEY=ak_… -- npx -y agentready-cloud-mcp
Works in Claude Code, Claude Desktop, Cursor, Windsurf and any MCP client:
✂️ tokencut_compress: long context loses greetings, hedges and filler before the model reads it. Code, URLs and numbers stay untouched.
📄 url_to_markdown: your agent reads the article, not the navbar. In our own usage 903K tokens of HTML became 14K of Markdown.
🧱 json_shrink: API responses become TOON. 30 order objects went from 1,106 to 444 tokens.
🔌 Building an app? Run the gateway yourself: npx agentready-cloud-gateway, then point your OpenAI or Anthropic SDK at localhost. Only message text is sent to us for compression; your provider key and the answers never leave your machine. Hacker News asked for exactly this, thank you nateb2022 and nevon.
Your comments built this release:
“Can I see what was trimmed?” Every response reports tokens before and after, and the dashboard has a shareable savings receipt.
“What about output tokens?” The output lever (lite / caveman / ultra) makes the model answer tersely.
Free during the beta. There's a small box in the dashboard asking what you'd pay: your answer sets the price.
Tell me: which assistant should we support next, and where do your tokens go: context, web pages, JSON or answers?
I'll be here all day. Also on X: x.com/christalingx 🚀