# Connect Bestie

Connect Bestie Check to Claude, ChatGPT, Codex, Cursor, Grok or your own agent. Sign in once, or use a key from your account.

Bestie Check works where your agent already works. Your agent hands over a memo or report, reads the helicopter view (the claims it makes and the assumptions they rest on, free), and orders a full check when you say so. Every paid check starts with a quote you approve.

## One address for every tool

Any tool that supports remote MCP connects to `https://www.hirebestie.com/api/check/mcp`. The first time, your tool opens a Bestie sign-in page. After that it keeps working until you disconnect it.

## By tool

### Claude Code

```
claude mcp add --transport http bestie-check https://www.hirebestie.com/api/check/mcp
```

Then run `/mcp` inside Claude Code and choose Bestie Check to sign in.

### Claude on the web and desktop

Open your connector settings, add a custom connector, and paste the address above. Sign in when asked.

### ChatGPT

Add a custom connector in your ChatGPT settings (developer mode, on plans that offer it) and paste the address above.

### Codex

```
[mcp_servers.bestie-check]
url = "https://www.hirebestie.com/api/check/mcp"
```

Add this to your Codex configuration file, then sign in when Codex asks.

### Cursor

```
{
  "mcpServers": {
    "bestie-check": { "url": "https://www.hirebestie.com/api/check/mcp" }
  }
}
```

Save it as `.cursor/mcp.json` in your project, or in your user settings to use it everywhere.

### Grok

Add a custom connector in Grok's connector settings and paste the address above.

## For scripts and other agents: a key

When there is no browser to sign in with, make a key at [your account](/account/api-keys). One key opens the command line, the API and MCP. It is shown once, it can be revoked at any time, and it expires on the date you pick.

```
export BESTIE_API_KEY=<your key>
npx -y @mybestie/check capabilities
npx -y @mybestie/check prepare --file memo.pdf --wait
```

To sign in from a terminal instead, run `npx -y @mybestie/check login`. It opens your browser and stores the credential on your machine.

## What happens next

1. Your agent gives Bestie the document. Bestie reads it and returns the helicopter view: every claim, and every assumption each claim rests on. This step costs no units.
2. You choose what to check: the argument, your own standards, or both. Bestie returns a quote in outcome units.
3. You approve the quote. Nothing is charged before that, and a check that cannot run costs nothing.
4. Findings come back with the place in the document, and a short list of what to change. Revise, then check again.

> Agents work on an account that already has units: add units to your Bestie account on the website first (plans are on the [pricing page](/pricing)). The connector never sells units and an agent never buys them. Your agent can only spend what you approve, from your balance, and an account with no units gets a plain message saying so; nothing runs and nothing is charged.

## If something does not work

- The sign-in page does not open: copy the address into your tool's connector settings again and retry, or use a key.
- A tool is missing from your list: disconnect and reconnect so the tool asks for the current list.
- A key stops working: check its expiry on your account page, or make a new one.
- Anything else: use [Report a problem](/check) in the app, or write to us from the [contact page](/contact).

Tool reference: [MCP](/docs/mcp) and [CLI](/docs/cli). The HTTP API description is at [/api/check/openapi.json](/api/check/openapi.json).