Free trial, then Pro from $6.67 a month.

Try it free

Guide4 min read

How to use imread: on the website and with AI agents

nglsongChecked against the code as it ships
On this page5 sections
  1. 1On the website
  2. 2With an AI agent: MCP
  3. 3With a script: the HTTP API
  4. 4With a prefix: Jina Reader style
  5. 5For agent builders

Paste a link into imread and you get the main text of the page, without menus, ads or scripts. Reading, copying and listening are free and need no account. imread Pro adds Studio voice, Pro translation and a faster MCP for agents.

On the website

  • Open imread.org and paste a link into the field, or tap Paste. A pasted link opens at once; a typed one opens on Enter. Share text such as Title https://… works: the link is picked out.
  • The article opens in the Text view. Switch to Markdown to keep headings, lists, tables, links and code.
  • Tap Copy to copy plain text or Markdown in one click.
  • Paste plain text instead of a link to listen to it or translate it. Translate and read aloud also work on any opened article.
  • Every result has its own address, https://imread.org/?url=<link>, so you can share it or refresh it.

Tables are kept: tab-separated in Text, so they paste into a spreadsheet as columns, and as pipe tables in Markdown. WeChat Official Account articles, Chinese and Japanese pages in legacy encodings, and SEC EDGAR filings are supported. Pages that build their text with JavaScript, need a login or block automated requests can't be read.

With an AI agent: MCP

The MCP server at https://imread.org/api/mcp has one tool, extract_article (url, optional format, offset, max_chars). No key. In Claude Code:

claude mcp add --transport http imread https://imread.org/api/mcp

Then ask: "Read https://example.com/post with imread and summarize it." A long article arrives in parts of about 20,000 tokens; each part ends with offset=N for the next one, or with [End of the article.]. Setups for Codex, Cursor, VS Code and Claude Desktop are in the MCP guide and on the MCP page.

With a script: the HTTP API

curl "https://imread.org/api/extract?url=https://example.com/post&format=markdown"
curl "https://imread.org/api/extract?url=https://example.com/post&format=text"
curl -X POST https://imread.org/api/extract -H "Content-Type: application/json" \
  -d '{"url":"https://example.com/post","format":"markdown"}'

Without format the answer is JSON with title, byline, siteName, lang, words, markdown and text. With a text format, offset and max_chars fetch a long article in parts. Errors come back as a readable sentence. The limit is 30 pages a minute per caller; an imread Pro key raises it to 600.

With a prefix: Jina Reader style

Put imread in front of the link:

curl https://imread.org/https://example.com/post
curl -H "Accept: application/json" https://imread.org/https://example.com/post
curl -H "X-Respond-With: text" https://imread.org/https://example.com/post

The plain answer has Title:, URL Source: and Markdown Content:. X-Respond-With takes markdown (default) or text; other values are a 400. Coming from Jina? See the comparison.

For agent builders

Everything is machine-readable: the API docs (send Accept: text/markdown for Markdown), /openapi.json, /.well-known/api-catalog, the MCP server card and an agent skill under /.well-known/. Pages are not stored: each request fetches the page afresh.

All posts →