The Lyrenth MCP server lets you read web pages through Lyrenth's index in any MCP client. It is designed to provide AI agents with clean, token-efficient web content as an alternative to raw HTML.
Purpose Convert public web pages into a unified AIDocument format: stable Markdown body text, plus title, description, and page structure, while stripping out navigation bars and boilerplate content. Agents read cleaned, low-token content, and each result also shows how many tokens were saved compared with the raw page.
Main capabilities The server exposes three tools:
- read_url: converts a single public web page into a clean AIDocument.
- read_urls: performs the same operation on up to 20 URLs in a single batch call.
- check_usage: reports your plan tier and quota usage. Reads are resolved through Lyrenth's cross-call cache; for verified domains, the canonical version authored by the publisher is returned and kept up to date by their change signals. Each tool explicitly declares all four MCP tool annotations as booleans (read-only, non-destructive, idempotent, open or closed world), and the HTTP transport comes with a set of 14 tests covering its security posture.
How to use Hosted, no installation: if your MCP client supports remote HTTP servers with custom request headers (Claude and Cursor remote connectors support this), you do not need to run anything: URL: https://api.lyrenth.com/mcp Header: Authorization: Bearer aiwk_your_key_here The three tools and the key are the same. The npx path is suitable for clients that use MCP over stdio, as well as users who prefer a local process. Get a free API key at https://lyrenth.com/signup (2,000 reads per month, no credit card required). Add the server to your MCP client:
- Claude Desktop (one-click): download lyrenth-mcp.mcpb from the latest release and open it. Claude Desktop will install the server and ask for your API key, with no need to edit configuration files. The bundle carries the same code as the npm package. To build it yourself, run: ./scripts/build-mcpb.sh .
- Claude Desktop / Cursor (manual configuration): add it to your MCP configuration (claude_desktop_config.json for Claude Desktop). Then ask the assistant to read a page, for example: "Read https://example.com/article and summarize it." It will call read_url and return the cleaned AIDocument.
Why read through Lyrenth
- Cleaner and cheaper: one stable AIDocument shape per URL; far fewer tokens are sent to the model than with raw HTML.
- Cross-call cache: when the same URL is fetched by multiple agents, it converges to the minimum number of origin fetches, so it is fast and origin-friendly.
- Canonical version returned after verification: when the site owner has verified on Lyrenth, you get the version they authored, kept fresh by their change signals.
Privacy The server sends only two things to Lyrenth's API (api.lyrenth.com): the URL you ask it to read, and the API key used to authenticate and meter the call. Nothing else leaves your machine: no page content you hold locally, no conversation context, no telemetry. For how Lyrenth handles crawled pages and account data, see the privacy policy:https://www.lyrenth.com/privacy。
License MIT. See LICENSE. Part of the Lyrenth project. The AIDocument format is an open contract; see https://lyrenth.com/llms-full.txt。
Limitations The materials do not provide pricing for plans other than the free quota, performance metrics, or service-level information; details such as the specific range of client compatibility and rate limits are not provided.
