> ## Documentation Index
> Fetch the complete documentation index at: https://docs.getmcpulse.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Connect

> Point Claude, Cursor or any MCP client at MCPulse over streamable HTTP.

The endpoint is:

```
https://api.getmcpulse.com/mcp
```

Authentication is OAuth — there is no key to paste. Your client discovers where to sign in, opens a browser, and you approve on a consent screen.

## Claude

Add MCPulse as a remote MCP server in your Claude settings, using the URL above. Claude will prompt you to authenticate; sign in with your MCPulse account and approve.

## Claude Code

```bash theme={null}
claude mcp add --transport http mcpulse https://api.getmcpulse.com/mcp
```

Then run `/mcp` inside Claude Code to start the OAuth flow.

## Cursor, VS Code, Windsurf and others

Any client supporting **streamable HTTP** and **OAuth 2.1** works. In a JSON config it usually looks like this:

```json theme={null}
{
  "mcpServers": {
    "mcpulse": {
      "type": "http",
      "url": "https://api.getmcpulse.com/mcp"
    }
  }
}
```

Consult your client's documentation for the exact key names — they differ.

<Note>
  There is no stdio option. MCPulse is a hosted service, not a local process, and OAuth needs a browser.
</Note>

## What approving does

You are sent to a consent screen in the MCPulse dashboard. It is the only thing on the page — no sidebar, no navigation — because a screen whose job is one irreversible decision should not offer somewhere else to click.

Approving issues the client a scoped access token and a refresh token. The client can then read your account's analytics, and nothing else. See [Authentication](/mcp/auth).

## Check it worked

Ask your agent to list your servers. It should call `list_mcps` and come back with your MCPs by name.

```
What MCP servers do I have on MCPulse?
```

If it reports it cannot authenticate, see [Troubleshooting](/mcp/troubleshooting).

## Naming your MCPs helps

The tools accept an MCP **by name** as well as by id, matched case-insensitively — so "how is the orders server doing" works without a `list_mcps` round trip first.

Two MCPs with the same name are ambiguous, and the server refuses to guess rather than reporting one server's numbers under the other's name. Distinct names are worth keeping.

## Related

* [Playbooks](/mcp/playbooks)
* [Tools](/mcp/tools)