# Connect an AI agent

> For the complete documentation index, see [llms.txt](https://crawlcoach.com/docs/llms.txt).

Add the CrawlCoach MCP server to Claude Code, Codex, Claude, VS Code, or Zed. Your agent signs in with your CrawlCoach account and sees only your own sites.

The CrawlCoach MCP server is a remote server that uses the Streamable HTTP transport. Your agent connects to this address:

```text
https://crawlcoach.com/mcp
```

Every plan includes the MCP server. Fix lists and rechecks come with a paid plan.

The first time your agent calls a tool, it asks you to sign in to CrawlCoach. Sign in and click **Allow**. You need a CrawlCoach account first: see [Get started](https://crawlcoach.com/docs/get-started).

## Claude Code

Claude Code changes your site's code for you. To connect it, follow these steps:

1. In your site's project, run the following command:

   ```bash
   claude mcp add --transport http crawlcoach https://crawlcoach.com/mcp
   ```

2. Type `/mcp`, choose **crawlcoach**, and choose **Authenticate**.
3. Sign in to CrawlCoach and click **Allow**.

## Codex

Codex changes your site's code for you. To connect it, follow these steps:

1. Run the following command:

   ```bash
   codex mcp add crawlcoach --url https://crawlcoach.com/mcp
   ```

2. In the sign-in page that Codex opens, sign in to CrawlCoach and click **Allow**.

If the sign-in page doesn't open, run `codex mcp login crawlcoach`.

## Claude on the web and the desktop app

To add CrawlCoach as a connector, follow these steps:

1. Open **Customize > Connectors**.
2. Click **Add**, and then click **Add custom connector**.
3. Enter `CrawlCoach` as the name and `https://crawlcoach.com/mcp` as the address, and then click **Continue**.
4. Click **Add**, and then click **Connect**.
5. Sign in to CrawlCoach and click **Allow**.

Claude can't change your site's files. It reads your fix list and writes each change for you to paste. For an agent that changes your code, use Claude Code or Codex.

## VS Code and Zed

VS Code with GitHub Copilot and Zed connect to remote MCP servers that use OAuth. Add a remote MCP server with the address `https://crawlcoach.com/mcp`, and then sign in to CrawlCoach and click **Allow**.

Cursor, Windsurf, and Gemini CLI aren't supported.

## Check the connection

Ask your agent to list your CrawlCoach sites. It calls `list_sites` and shows each site with its page counts.

## Give your agent the fix prompt

The fix prompt tells your agent how to work through your fix list. To copy it with your site's name in it, open your site in CrawlCoach and click **Copy prompt for Claude or Codex**. Then paste it into your agent in your site's project.

Agents that support MCP prompts can also use the `fix_my_site` prompt. For details, see [MCP tools reference](https://crawlcoach.com/docs/mcp-tools).

## Permissions

When you click **Allow**, your agent gets these scopes:

- `crawlcoach:read`: read your sites, fix lists, and pages.
- `crawlcoach:write`: ask CrawlCoach to recheck pages.

Your agent sees only the sites in your own CrawlCoach account. It can't change your website, your account, or your billing.

## Disconnect your agent

To disconnect, do one of the following:

- In Claude Code, run `claude mcp remove crawlcoach`.
- In Codex, run `codex mcp logout crawlcoach`.
- In Claude, remove the CrawlCoach connector.

## Troubleshoot

If your agent can't find the CrawlCoach tools, sign in again:

- In Claude Code, type `/mcp`, choose **crawlcoach**, and choose **Authenticate**.
- In Codex, run `codex mcp login crawlcoach`.
- In Claude, open the CrawlCoach connector and click **Connect**.

If the sign-in succeeds but every tool fails with `ACCOUNT_REQUIRED`, create a CrawlCoach account with the same Google account first.