Universal content reader · Agent Skill + CLI + MCP

Turn a URL into source content your AI agent can actually use.

x-reader is a universal content-reading layer for AI agents. It routes X/Twitter, articles, YouTube, Bilibili, podcasts, WeChat, Xiaohongshu, Telegram, RSS and open-web URLs through the right read path, then returns normalized source content for downstream research, summarization and automation.

One reader layer, many sources

Stop wiring a different scraper into every agent workflow.

Give x-reader the URL. It detects the source type, chooses the appropriate fetcher or fallback, and produces a common structured result instead of making every agent understand every platform itself.

X / TwitterArticlesYouTubeBilibiliPodcastsWeChatXiaohongshuTelegramRSSOpen web
Core data path
1 · InputURL
2 · Routeplatform fetcher
3 · NormalizeUnifiedContent
4 · ConsumeAgent / JSON / MCP

When attached audio or video matters, the workflow can continue through subtitles or transcription instead of pretending surrounding metadata is the media itself.

What the product actually does

Source ingestion first. Evidence awareness second.

Evidence Receipt is a guardrail on top of the reader, not the product definition. The primary job is to retrieve and normalize the source so an agent can use it.

01
Retrieve the source

Platform-specific fetchers and fallbacks turn a user-supplied URL into actual source content.

02
Normalize for agents

Title, URL, source type, body content and relevant metadata become one structured payload.

03
Complete material media

If video or audio matters, x-reader can continue through subtitles, transcripts or local media tooling instead of stopping at page metadata.

Evidence-aware by design

It also tells the agent when the read is incomplete.

This is the trust layer: x-reader does not substitute a search snippet for the source, a post caption for its attached video, or a video description for spoken content.

PASSThe source layers needed for the task were actually read.
PARTIALUseful source content was retrieved, but a material layer is still missing.
FAILThe requested source could not be fetched reliably.
UNKNOWNThe returned material is too ambiguous to prove coverage.
Four ways to use the same reader

Web for a quick read. Skill, CLI and MCP for the full agent workflow.

Easiest · let your AI install it

Paste this into Claude Code, Codex or Cursor:

Set up x-reader on this computer by following the official instructions at https://github.com/runesleo/x-reader/blob/main/install.md.
  1. Copy the sentence above
  2. Paste it into the AI you already use
  3. It installs x-reader, runs the read-only x-reader doctor self-check and reads a first link with you. Anything that needs a login or a key, it tells you how to do yourself.
Hosted subsetWeb

Public X, YouTube and generic web URLs. Zero install; useful for trying the reader and inspecting an Evidence Receipt.

Open hosted app →
Recommended for agentsAgent Skill

Use x-reader directly when an agent receives a link and needs original content or source-grounded analysis.

npx skills add runesleo/x-reader --skill x-reader
For scripts & pipelinesCLI

Single or batch URL reads with complete machine-readable output.

x-reader URL --json
For toolchainsMCP

Expose read_url, read_batch, list_inbox and detect_platform as reusable agent tools.

MCP setup →
Full x-reader

More than the hosted webpage.

The full local/agent workflow covers sources that may require platform APIs, browser fallback, saved local sessions, subtitles or transcription. The hosted beta deliberately exposes only the public-safe subset.

URL → detect platform → fetch / fallback → media completion when needed → UnifiedContent → agent
What it is not

Not another summarizer. Not just an AI lie detector.

Summaries and analysis happen after reading. Evidence status is there to keep the source boundary honest. The core product is the reusable reader layer that gets source content into the agent in the first place.

What it won't do

Read-only, within bounds, honest about gaps.

  • Read-only. It never posts, likes, or logs in to an account for you.
  • No made-up coverage. A search snippet is not the source, and an unread video is never summarized — gaps stay PARTIAL or FAIL.
  • Your keys, your call. The local version keeps cookies and keys on your computer and only sends them to a provider you explicitly enable (for example Groq for transcription); the web version reads public links only, no login.
  • Privacy of hosted usage counts. If enabled, anonymous daily page-request and read-status aggregates are retained for up to 90 days; no submitted URLs, IPs, cookies or source content enter these counters.
  • Pages are data, not orders. Retrieved pages, posts and subtitles are treated as material, never as instructions to your AI.
The simplest mental model

Give your agent a URL. Let x-reader handle the source.

One ingestion layer across platforms, structured output for agents, and explicit coverage when something is still missing.