It reads every page
On its own, an assistant estimates from the pages it opens — or, in Claude Code, writes a one-off crawler that counts the menu on every page. Word Counter follows every link and the sitemap until there is nothing left to read.
Connect Word Counter to your AI assistant. When you ask about a website, it crawls every page and comes back with the real total — the same count as our reports, not a guess from the homepage.
Free with a Word Counter account. Works with Claude Code, Claude, ChatGPT, Cursor, VS Code and any app that speaks MCP.
I counted every page with Word Counter:
2 696 246 words on 2 494 pages
39% of it repeats text from other pages; 1 604 010 words are new.
Full report, every page →
On its own, an assistant estimates from the pages it opens — or, in Claude Code, writes a one-off crawler that counts the menu on every page. Word Counter follows every link and the sitemap until there is nothing left to read.
The main text a visitor reads — without menus, footers and forms, and with each duplicate page counted once.
An estimate of the sentences repeated across pages, and the split by language, section and subdomain — the numbers a quote is built on.
Sign in with Google and make a connection. You get a personal URL — and a key for apps that take one — shown once.
Get your connection URLPick your app below and paste it in. It takes a minute.
“How many words does example.com have?” Your assistant starts the count, follows it, and answers with the total.
In a terminal, add the server with your key in a header:
claude mcp add --transport http word-counter https://markdown.wxrks.app/mcp \
--header "Authorization: Bearer wck_your_key"Add --scope user to have it in every project. Then type /mcp in Claude Code to check that word-counter is connected, and ask away.
https://markdown.wxrks.app/mcp/wck_your_keyLeave the advanced settings empty and add it. In a chat, make sure Word Counter is switched on in the tools menu. On the Free plan Claude allows one custom connector.
https://markdown.wxrks.app/mcp/wck_your_keyCustom MCP servers depend on your ChatGPT plan and, in a workspace, on what your admin allows.
Cursor — in ~/.cursor/mcp.json:
{
"mcpServers": {
"word-counter": {
"url": "https://markdown.wxrks.app/mcp",
"headers": { "Authorization": "Bearer wck_your_key" }
}
}
}VS Code — in .vscode/mcp.json:
{
"servers": {
"word-counter": {
"type": "http",
"url": "https://markdown.wxrks.app/mcp",
"headers": { "Authorization": "Bearer wck_your_key" }
}
}
}Any other app: use Streamable HTTP with the server URL and the key as a Bearer token — or, if it only takes a URL, your personal URL.
Your assistant picks these itself; you just ask in your own words.
| Tool | What it does |
|---|---|
count_words | Counts a website: the total, pages, repetition, and the split by language, section and subdomain, with a link to the full report. The whole domain by default, or one host only. Reuses an earlier count of the same site unless asked to count again. |
get_word_count | Follows a count that’s still running, and returns the result when it’s done. |
continue_word_count | Carries a large site past its estimate to the very last page. |
list_word_counts | The sites you’ve counted, with their totals. |
get_pages | The pages of a count with their words, largest first, optionally only one section. |
Seconds for a small site, minutes for a large one. Your assistant waits up to a minute, then checks back on the count until it’s done.
At about 50,000 words a count pauses with an estimate of the whole site. Ask your assistant to continue and it counts every page.
Into Your runs, like any count you start here — each with its report, every page and a CSV.
Nothing extra: the same as counting on the website. Pro counts run faster.
No — treat it like a password: anyone with it can count as you. You can revoke it at any time and make a new one.
Streamable HTTP at the URL below, plain JSON responses, no sessions. Protocol versions 2025-03-26 to 2026-07-28. Authentication: a Bearer key, or the key in the URL path.
https://markdown.wxrks.app/mcp