View Liveblocks Documentation

Purpose

Look up Liveblocks developer documentation for a given topic and return the canonical page title, its URL, and the full page content (concepts, installation steps, code snippets, API references). Liveblocks publishes its docs as machine-readable markdown alongside the rendered HTML, so the optimal path is a plain HTTP fetch — no browser, no JavaScript rendering, no authentication, and no anti-bot stealth. Read-only.

When to Use

Answering a developer question from authoritative Liveblocks docs (e.g. "how do I set up Comments in Next.js?", "what hooks does @liveblocks/react export?", "how does access-token auth work?").
Pulling a specific guide, API reference, or tutorial as clean markdown for summarization, RAG ingestion, or code generation.
Enumerating every available docs page (the llms.txt index is a complete URL list).
Bulk-ingesting the entire documentation corpus in a single request (llms-full.txt).

Workflow

Recommended method: fetch. Liveblocks (a Next.js site on Vercel) exposes its docs three machine-readable ways. None require a browser, proxies, verified sessions, cookies, or auth — a bare HTTP GET works. Pre-run probe and live testing on 2026-06-13 showed no anti-bot of any kind on liveblocks.io.

Discover the page URL from the index. Fetch the docs index, which lists every documentation page with its title and URL:
```
browse cloud fetch "https://liveblocks.io/llms.txt"
```
The index is grouped into sections (Readme, Ready-made features, SDKs and packages, Developers, Documentation). The ## Documentation section is an explicit TITLE → https://liveblocks.io/docs/... list and is prefixed with the note: "suffix docs URLs with .md to view markdown files." Match the requested topic to the closest entry (e.g. GET_STARTED_REACT → https://liveblocks.io/docs/get-started/react).
Fetch the page as markdown by appending .md to the docs URL:
```
browse cloud fetch "https://liveblocks.io/docs/get-started/react.md"
```
Returns Content-Type: text/markdown (served by the internal route /api/docs-raw/[...path]). The body starts with a YAML frontmatter block (meta.title, meta.parentTitle, meta.description) followed by the full markdown content, including fenced code blocks. Internal links inside the markdown are already rewritten with the .md suffix (e.g. [BlockNote with Next.js](/docs/get-started/nextjs-blocknote.md)), so you can crawl deeper by following them directly.
Extract and return. Parse meta.title from frontmatter for the page title, use the canonical (non-.md) URL as page_url, and return the markdown body (or a summary + extracted key_steps/code snippets) per the schema below.
(Optional) Whole-corpus fetch. When you don't yet know which page you need, or you're building a search index:
```
browse cloud fetch "https://liveblocks.io/llms-full.txt"   # ~2.9 MB, entire docs concatenated
```
Each page within is delimited by its own frontmatter block. Prefer the per-page .md fetch for targeted lookups — it's far smaller.

Browser fallback (only if the markdown endpoints ever change)

The .md / llms.txt endpoints are the contract; if they ever 404 or change shape, fall back to the rendered HTML:

browse open "https://liveblocks.io/docs/get-started/react" --remote   # bare session is fine
browse wait load --remote
browse wait timeout 2500 --remote          # Next.js hydration
browse get markdown body --remote          # or `browse get text body`

This works but is ~100× slower and more brittle (JS-rendered Next.js app). Use the markdown endpoint whenever it's available.

Site-Specific Gotchas

No anti-bot — do NOT enable stealth. The successful runs used a bare session with no --proxies and no --verified. The pre-run probe reported antibots: none detected and live fetches all returned 200. Turning on proxies/verified only adds cost and latency for zero benefit.
.md suffix is the killer feature. Appending .md to any https://liveblocks.io/docs/... URL returns the raw markdown source. This is stated directly in llms.txt. It is served by /api/docs-raw/[...path], not a static file.
404 on a .md URL returns HTTP 404 with a helpful body, not a silent redirect: "Sorry, we don't have a markdown file for this page. Try visiting the page directly... For an overview of all available documentation, see /llms.txt". If you hit this, your URL/path is wrong — re-derive it from llms.txt (titles in the index are UPPER_SNAKE aliases, not the page's real meta.title; trust the URL, not the alias).
Frontmatter is part of the body. The .md response begins with a --- YAML block (meta.title, meta.parentTitle, meta.description, sometimes alwaysShowAllNavigationLevels). Strip or parse it; don't treat it as page content.
Index titles ≠ page titles. In llms.txt the docs are labelled with screaming-snake aliases like GET_STARTED_REACT, FEATURES_COMMENTS_HOOKS, API_REFERENCE_REACT. The human page title comes from the fetched page's meta.title (e.g. GET_STARTED_REACT → "Get started with Liveblocks and React").
API reference pages are large. liveblocks-react.md is ~219 KB and llms-full.txt is ~2.9 MB. Budget for big responses; fetch the specific .md page rather than the full corpus when you can.
pricing.md exists too. Non-docs marketing pages aren't all markdown-backed, but /pricing.md is explicitly listed in llms.txt as a markdown alternative to /pricing.
Liveblocks also ships an MCP server (/docs/tools/mcp-server) and "agent skills" (/docs/tools/agent-skills) — but those are for operating Liveblocks projects/APIs, not for reading docs. For pure documentation lookup, the fetch path above is simpler and needs no setup.
Caching: responses are served via Vercel edge with X-Vercel-Cache: HIT and a long Age. Content is effectively static between releases; safe to cache locally.

Expected Output

// Success — single page fetched as markdown
{
  "success": true,
  "method": "fetch",
  "topic": "get started with Liveblocks in React",
  "page_title": "Get started with Liveblocks and React",
  "page_url": "https://liveblocks.io/docs/get-started/react",
  "content_summary": "Step-by-step quickstart for adding realtime collaboration to a React app using @liveblocks/client and @liveblocks/react: install packages, init the config file, set up LiveblocksProvider + RoomProvider, join a room with ClientSideSuspense, sync shared state with useStorage/useMutation, show presence with useOthers, and configure auth.",
  "key_steps": [
    "Install packages: npm install @liveblocks/client @liveblocks/react",
    "Initialize config: npx create-liveblocks-app@latest --init --framework react",
    "Set up LiveblocksProvider (publicApiKey) and RoomProvider (room id) in App.tsx",
    "Wrap room children in ClientSideSuspense for loading states",
    "Define typed Storage and set initialStorage on RoomProvider",
    "Read shared state with useStorage; update it with useMutation",
    "Access connected users with useOthers to build presence (e.g. avatar stack)",
    "Optionally add an authentication endpoint to restrict room access"
  ],
  "markdown": "---\nmeta:\n  title: \"Get started with Liveblocks and React\"\n  ...\n---\n\n# Get started ...",
  "error_reasoning": null
}

// Topic not matched to a page (index lookup found no good candidate)
{
  "success": false,
  "method": "fetch",
  "topic": "<requested topic>",
  "error_reasoning": "No documentation page in /llms.txt matched the topic. Closest candidates: [GET_STARTED_REACT, GET_STARTED_NEXTJS]. Ask the user to narrow the topic or pick a candidate."
}

// .md endpoint 404 (wrong path)
{
  "success": false,
  "method": "fetch",
  "topic": "<requested topic>",
  "page_url": "https://liveblocks.io/docs/<bad-path>",
  "error_reasoning": "GET /docs/<bad-path>.md returned 404 ('no markdown file for this page'). Re-derive the URL from /llms.txt."
}

view-documentation