Connect Haven to ChatGPT or Claude

You can connect your Haven account to ChatGPT, Claude, or any other AI assistant that supports remote MCP servers. Once connected, you can ask the assistant about your listings, calendar, rates, bookings, and guest conversations in plain language — and, if you allow it, let the assistant make changes such as blocking dates or updating rates.

The connection is an ordinary Haven authorization, the same kind a property management system uses. You sign in, you read a consent screen, and you can disconnect it at any time from Connected apps.

Connecting your account does not put your listing in front of travelers who ask ChatGPT for a place to stay. Guests never connect anything. How a public listing gets into those answers is covered in How guests find your stay in ChatGPT.

When to connect an assistant

• You want to ask questions across all your listings at once — "which weeks in October are still open", "what did I earn on the beach house last quarter" — without building a report.
• You want an assistant to draft work from your real data instead of guessing.
• You want to give an assistant permission to make routine changes (block dates, adjust a rate, set a minimum stay) while you supervise.
• You want a read-only assistant that can look but never change anything. That is what a Viewer gets automatically — see below.

The Haven MCP URL

Every assistant asks for the same thing: one address.

https://www.bookwithhaven.com/api/public/mcp

That is the whole configuration. You do not need an API key, a client ID, or a client secret. The assistant discovers how to sign you in from the URL itself, registers itself with Haven automatically, and sends you to Haven's own sign-in and consent screen.

Connect in Claude

1. Open Customize → Connectors (or go directly to claude.ai/settings/connectors). It is under Customize, not Settings.
2. Click the + button next to Connectors, then choose Add custom connector.
3. Enter a name — Haven — and paste the MCP URL above.
4. Leave the Advanced settings (OAuth Client ID and Secret) empty. Haven issues those automatically.
5. Click Add, then Connect. Claude opens Haven's sign-in page.
6. Sign in to Haven, read the consent screen, and click Allow access.
7. In a conversation, open the + menu in the composer and switch the Haven connector on. A connector that is added but not switched on for that chat cannot be used.

On Team and Enterprise plans, members cannot add a custom connector themselves. An owner adds it once under Organization settings → Connectors, and then each member connects their own Haven account from Customize → Connectors. That split is worth keeping: each person's connection is capped by their own Haven role.

Connect in ChatGPT

Today the way in is a custom connector, described below. A Book With Haven for Hosts app is on its way to ChatGPT's app directory; once it is listed, installing it from there (Apps → search Book With Haven for Hosts → Connect) replaces these steps, with no Developer mode and no URL to paste. It is the same Haven server with the same permissions and the same consent screen either way. This article will say so when the listing is live.

1. Click your profile → Settings → Apps & Connectors.
2. Scroll to the bottom, open Advanced settings, and turn Developer mode on. Without it, ChatGPT only sees Haven's search and fetch tools, not the full set.
3. Go back to Apps & Connectors and click Create.
4. Fill in a name (Haven), a short description, and paste the MCP URL above as the MCP server URL.
5. Choose OAuth as the authentication method — ChatGPT usually detects this from Haven automatically.
6. Tick I trust this application and click Create. ChatGPT opens Haven's sign-in page.
7. Sign in to Haven, read the consent screen, and click Allow access.
8. Start a new chat, open the connector picker in the composer, and enable Haven. Chats that were already open will not pick up a new connector.

Custom connectors in ChatGPT require a paid plan, and on Business or Enterprise workspaces an admin may need to allow custom connectors first.

What the consent screen means

Haven's consent screen is the only part of this flow Haven controls, and it is where the actual decision happens. Read it rather than clicking through it.

• The app's name, description, and logo at the top are whatever the application chose for itself. They are not a Haven endorsement.
• "Connecting as" names the Haven account being authorized. It comes from the session you are signed in to, never from the link you clicked. If it names an account you did not expect, cancel.
• "After you allow this, you will be sent to…" names the site the connection hands off to. For Claude that is an Anthropic domain; for ChatGPT, an OpenAI one. If it names somewhere you have never heard of, that is the signal to stop.
• "This covers every listing on this account, including listings you add later." There is no per-listing picker. A connection is account-wide.
• "It will be able to read" and "It will be able to change" are the two halves that matter. Each line is written in plain language — "See your bookings, their dates, status and totals" — and grouped by area. If the "change" section says Nothing. This connection can read only, the connection cannot alter anything on your account.
• "What it can never do" is fixed and true of every connection, no matter what was requested: it can never see or change your Haven subscription, card, or invoices; never move money beyond what you explicitly allowed; never add, remove, or change who has access to your account; never create or delete workspaces or close your account; and never sign in as you elsewhere in Haven.
• If you are reconnecting an app you have connected before, the screen also shows what is new, what is already allowed, and what will be removed — so a connection that quietly asks for less is visible, not silent.

The choice is allow or cancel. There are no per-permission checkboxes, because a half-granted connection fails in ways the app cannot explain to you.

The "Haven has not reviewed this application" banner

When you connect ChatGPT or Claude, you will see an amber banner directly under the app's name:

Haven has not reviewed this application It registered itself automatically, which is how AI assistants and developer tools connect. Its name, description and website are whatever it chose for itself. Only continue if you started this connection yourself, from an application you trust.

This is expected, and it is not a warning that something has gone wrong. AI assistants are handed a URL by you and register themselves with Haven on the spot — nobody at Haven vetted the application first, because there was nothing to vet before you pasted the URL. Partners that go through Haven's review process do not show this banner.

What the banner is really asking is one question: did you start this? If the consent screen appeared while you were adding a connector in ChatGPT or Claude, that is exactly right. If it appeared because you clicked a link in an email or a message, close it and start again from inside the assistant.

If your role is Viewer, the connection is read-only

Haven never lets a connection do more than the person who approved it can do.

If your role on the account is Viewer, the assistant's request for permission to make changes is dropped before the screen even renders. You will see:

• "It will be able to change — Nothing. This connection can read only."
• A section headed "Not included, because your role is Viewer", listing by name each change permission the app asked for and did not get.

This is deliberate rather than a failure. The assistant asks for everything Haven publishes, because nobody chose a narrower list on its behalf — so refusing the whole request outright would mean no Viewer could connect Claude or ChatGPT at all, just to read listings they can already see on their own dashboard. Instead the connection is narrowed to reads and you are told exactly what was dropped, so you can explain it if the assistant later says it cannot make a change.

The rule holds afterwards too. The role is re-checked on every single request, so a manager who connected an assistant and was later changed to Viewer stops handing that assistant write access immediately — no reconnection needed. Going the other way, a Viewer who is later promoted does not silently gain write access; widening a connection requires approving it again.

Viewers also cannot disconnect an application. They can see the list; removing an app is an owner or manager action.

Good to know

• One connection, one Haven account. The account is fixed at the moment you approve. If you manage several Haven accounts, each one is its own connection and its own consent.
• Money-moving actions take two steps. A refund or a price change is refused on the first attempt and has to be repeated with a confirmation token that is valid for five minutes and only for those exact amounts. An assistant cannot push a refund through on a single call.
• PMS locks still apply. If a connected property management system owns your rates or calendar for a listing, an assistant's attempt to change them is refused, exactly as it would be in the Haven dashboard. See PMS Integrations Overview.
• Sending guest messages is a separate permission from reading them, and both are shown separately on the consent screen.
• If someone on your team connects an assistant on the owner's behalf, the consent screen says so and the account owner is emailed that it happened.
• Disconnecting is immediate, from bookwithhaven.com/connected-apps.

Troubleshooting

The assistant says it has no Haven tools. In ChatGPT, confirm Developer mode is on and that you started a new chat after creating the connector. In Claude, confirm the connector is switched on in the + menu for that conversation.

The assistant can read but not change anything. Either your role is Viewer, or you approved a read-only connection. Open Connected apps, expand What it can do, and check. Widening it means an owner or manager disconnecting the app and connecting it again from their own Haven sign-in; a Viewer cannot do either half.

A change was refused. The two common reasons are a PMS that owns that part of the listing, and a permission that was never granted. Both come back as a plain-language refusal the assistant can show you.

You did not start the consent screen you are looking at. Cancel it. Nothing is granted until you click Allow access.

Related Articles

• How guests find your stay in ChatGPT
• Connected apps
• Workspaces and Team Access
• PMS Integrations Overview