live fetch · no training cutoff

Fresh context for coding agents.

The fetch layer between your coding agent and the real, current source. Library docs, live pages, and structured data, pulled on demand instead of recalled from a training cutoff.

$curl -fsSL getctx.sh/install.sh | sh

Read the docs →

ctx · docsreact.dev

$ ctx docs react "useEffect cleanup"

# /reactjs/react.dev

## 1. [<StrictMode>](github://reactjs/react.dev@main/.../StrictMode.md)

- The `useEffect` hook with a proper cleanup function added to

disconnect from the chat, resolving accumulating connections.

## 2. [Synchronizing with Effects](github://reactjs/react.dev@main/.../synchronizing-with-effects.md)

- Adding a cleanup function to the `useEffect` to disconnect when

the component unmounts or before the effect re-runs.

## 3. [useEffect](github://reactjs/react.dev@main/.../react/useEffect.md)

- A symmetrical setup and cleanup example, where cleanup undoes setup.

---

Use `ctx read <url>` to fetch full documents.

backed by Context7's live indexruns in Claude Code, Cursor, Windsurf, ZedMCP server + skill

Your agent's knowledge has an expiry date.

Models are frozen at a training cutoff. The day a library ships a new API, your agent keeps writing the old one, with total confidence.

without ctx

$ "configure vite 8"

export default defineConfig({
  optimizeDeps: { disabled: true }, // removed in v5
})

✗ confidently wrong — no warning, no doubt

with ctx

$ ctx docs vite "config"

# /vitejs/vite

## 1. [Configuring Vite](github://vitejs/vite@main/docs/config/index.md)
- The most basic config file structure.
- How to explicitly specify a config file using the CLI.

## 2. [Plugin API](github://vitejs/vite@main/docs/guide/api-plugin.md)
- Return a partial config from the `config` hook; Vite deep-merges it.

Eight commands. One fetch layer.

Every command returns agent-ready text or data. Run them from your shell, your editor, or any MCP client.

$ ctx --help

search<library>Resolve a package name to its indexed docs on Context7.docs<library> <topic>Pull the doc sources for a library, narrowed to a topic.read<url>Turn any page (https:// or github://) into clean markdown.screenshot<url>Render a full page to an image via headless browser.links<url>List every link on a page for the agent to follow.scrape<url> <selector>Pull elements matching a CSS selector, structured for parsing.json<url> --schemaExtract typed JSON from a page against a schema, with AI.crawl<url> [--limit]Walk a documentation site and collect every page.

Watch it work.

This is how an agent actually uses ctx: search to find the library, docs to scope the topic, read to pull the full document. Real output, no narration.

ctx · playgroundready

# watch an agent run the ctx workflow

open the playground

ctx searchfind the right library

ctx docsscope to the topic

ctx readpull the full document

today docs from Context7's live index, pages through a headless browser.soon a first-party getctx API, so nothing is stitched.

Free to start. An API when you outgrow it.

The CLI is free for individual use. A first-party API with higher limits and team keys is on the way.

ctx CLIfree

✓ all eight commands
✓ docs, read, screenshot, scrape, crawl
✓ MCP server + Claude Code skill
✓ Context7 library index

$ curl -fsSL getctx.sh/install.sh | sh

getctx APIsoon

○ first-party endpoints, no stitching
○ higher rate limits + edge caching
○ team keys + usage dashboards

Join the waitlist →

Give your agent a source of truth.

Install in ten seconds. No signup.

$curl -fsSL getctx.sh/install.sh | sh

or try it in the browser →