# SEO for Claude Code via MCP

> Run SEO audits in Claude Code via MCP. Add the Crawlwise server with one command, sign in with OAuth, and see what it will and will not do before relying on it.

Canonical: https://crawlwise.site/docs/claude-code

Crawlwise gives Claude Code SEO audits and AI crawler checks through MCP. The server speaks streamable HTTP at `https://crawlwise.site/mcp`. Adding it takes one command and signing in is OAuth, so no key has to be pasted anywhere.

## Add the server

```
claude mcp add --transport http crawlwise https://crawlwise.site/mcp
```

Then run `/mcp` inside Claude Code and authenticate when it prompts. The browser opens, you approve the scopes, and the tools appear for that session and every later one.

A person who would rather not use OAuth can paste an API key instead — [create one](https://crawlwise.site/account/api) (Growth and Agency) and pass it as a bearer token in the same command with`--header`.

## Or install the plugin: skill and MCP in one

The plugin adds the MCP server and the crawlwise skill, which teaches Claude the audit, fix, re-audit loop and the rule that only **priority** items are defects. Run these inside Claude Code:

```
/plugin marketplace add Adnanarodiya/crawlwise-skills
/plugin install crawlwise@crawlwise
```

Then ask: *“audit https://example.com with crawlwise and fix the priority items”*. The source is public at [Adnanarodiya/crawlwise-skills](https://github.com/Adnanarodiya/crawlwise-skills).

## Not on Growth yet? The skill and the CLI

The skill on its own installs from npm, and the CLI audits one page a day without an account:

```
npx crawlwise skills install
npx crawlwise audit https://example.com --format llm
```

The free audit returns the score and the top issues. Every fix, its evidence and Core Web Vitals need a key.

## Verify it worked

```
Ask Claude: "use crawlwise to check whether GPTBot can crawl example.com"
```

A working connection answers with a per-crawler verdict: allowed, blocked, challenged, or not established — and it will say “not established” rather than guessing when a fetch failed. If the list is empty, `/mcp` shows whether the server connected and what it reported.

## What it costs

Connecting needs Growth or Agency: every MCP call is made with your key, or with the key the sign-in creates. The checker tools — AI crawler access, schema validation, Core Web Vitals and the generators — spend no credits. A full audit starts at 1.25 credits; ask for estimate_cost first if you want the price before the audit starts. On the website the same checkers are free with no account at all.

## What it will not do

It will not invent a score, a percentage, or a readiness rating, because the service does not produce any. It cannot tell you whether Google will rank a page, and it will not present a checklist score as a Google score. When a value did not come back it says **not measured** — and a model reading that answer should not round it to zero.

[MCP reference](https://crawlwise.site/docs/mcp) · [What crawlwise cannot tell you](https://crawlwise.site/limits) · [Cursor](https://crawlwise.site/docs/cursor) · [Windsurf](https://crawlwise.site/docs/windsurf)