Wherever your agent
reads the web.
Lyrenth drops into your stack as one clean index layer for the web. Add it to an MCP client, call a one-line reader, or wire in a native SDK, and any URL comes back as a clean AIDocument instead of raw HTML.
A web reader your
agent just uses.
Lyrenth is listed in the extension directories inside Claude Desktop and Cursor, so you can install it from the app and never open a config file. Your agent gets a read_url tool and reads the web through Lyrenth's index, no glue code, no scraping.
Every other MCP client takes one line:
Plug in your way.
Same key, same clean AIDocument, whichever surface you reach for.
MCP server
A read_url tool for any MCP client. Install it from the directory inside Claude Desktop or Cursor, or point any other client at the package.
Reader endpoint
One authenticated GET returns clean Markdown. A shell one-liner, or any HTTP client.
Python SDK
A dependency-free client, plus LangChain and LlamaIndex adapters for your RAG pipeline.
TypeScript SDK
A dependency-free client, plus a ready-made tool for the Vercel AI SDK.
Already in your RAG stack.
Drop-in loaders and tools for the frameworks you already build on.
LangChain
LyrenthLoader drops clean Markdown documents straight into your chain or vector store.
pip install 'lyrenth[langchain]'LlamaIndex
LyrenthReader loads any URL as a Document, ready to index, chunk, and query.
pip install 'lyrenth[llamaindex]'Vercel AI SDK
A read_url tool the model can call mid-generation to pull a clean page into context.
npm install lyrenthStart from working code.
Every AI answer that reads the web pays for reading pages. These projects show how much less it costs when the reading is clean.
Eleven ready-to-use agents
One command each. Compare two products, brief a company, answer from documentation, pull several pages into a table. Every agent reads its pages through Lyrenth and answers with the page behind each claim.
Research agent
Give it a question and a list of URLs. It reads every page through Lyrenth, drops duplicates, stays inside a token budget and answers with numbered citations, using any OpenAI-compatible model you choose.
The web, indexed for AI.
Every read resolves through one clean index: rendered, cached, and consent-aware.
Clean, low-token content
JavaScript rendered when needed, navigation and boilerplate stripped. Far fewer tokens than raw HTML, and you get the whole page, clean and complete.
Cross-caller cache
The same URL read by many agents collapses to a minimal number of origin fetches. Fast for you, friendly to the site.
Consent-backed canonical
When a publisher has verified their domain, a page they change is fetched again at the front of our crawl queue, so you read the new version sooner.
Read the web,
the clean way.
Free tier: 2,000 reads / month, no card. One key works across every integration on this page.
Claude, Cursor, LangChain, LlamaIndex, Vercel, Python, and other product names, marks, and logos referenced on this page are the property of their respective owners. Shown for identification and compatibility only; no partnership or endorsement is implied.