Connect an AI assistant to UpTrust

You can connect an AI assistant such as Claude, ChatGPT, or Grok to your UpTrust account. Once connected, your assistant can use your trust network when it answers you. It can rank people or posts by how much your network trusts them, check whether your network trusts someone here, and find your most trusted voices on a topic. It can also set up an event and invite people for you, which we have only confirmed in Claude.

Where to find it: You start in your assistant's own settings, under Connectors. There is nothing to switch on inside UpTrust first, and you do not need an UpTrust account yet.

What you'll need

You need an assistant account that supports custom connectors. On Claude, ChatGPT, and Grok that means a paid plan today. If you do not see a Connectors section in your assistant's settings, your plan may not include them yet.

You do not need to sign up for UpTrust first. If you are new, you create your account in the UpTrust window that opens when you connect, as described below.

Claude and Grok use this address: https://mcp.uptrusting.com/mcp

Copy it exactly, ending in /mcp. The address without that last part is a different thing and the dialog will refuse it. ChatGPT has its own address, below.

Connect from Claude

1. Open your settings and find the Connectors section.
2. Choose to add a custom connector.
3. Paste the address above and confirm.

Connect from ChatGPT

1. Open Settings, then Connectors, then Advanced settings, and turn on Developer mode. Custom connectors sit behind this switch.
2. Back in Connectors, choose Create.
3. Name it UpTrust, paste https://mcp.uptrusting.com/chatgpt, and set Authentication to OAuth.

Developer mode lives on the ChatGPT website, not in its phone app. If you only want to send us your introduction, you can paste it in instead on any plan.

Connect from Grok

1. Open the Connectors page in Grok's settings.
2. Choose New Connector, then Custom.
3. Paste the address above and confirm.

Approve on UpTrust

The last part is the same everywhere. A browser window opens on UpTrust and shows which assistant is asking and what it is asking for.

If you are new to UpTrust

1. Click Continue with Google, Apple, or GitHub, then follow that service's own screen. Google and Apple create your UpTrust account with that email. GitHub makes a private account that nobody can see and that your assistant can only read with. If you would rather use none of them, choose Sign up with email at the bottom.
2. If you came from Claude, ChatGPT or Perplexity, that same click also approves the connection, so you are done. You see the approval screen next instead, where you read it and click Approve, if you came from any other assistant, if you signed up with email, if you were first asked to choose a username, or if the click signed you in to an account you already had.

Click Cancel instead if you change your mind. Nothing is connected.

If you already have an account

1. Click Log in at the top of the page, unless you are already logged in.
2. Read what the assistant is asking for, then click Approve.

Either way, go back to the assistant, open a chat and check its tools menu. Turn on UpTrust there if it is not already on. There are no keys or codes to copy. If your assistant will not connect, tell us on the feedback board and say which assistant you tried.

What the assistant asks permission for

The consent screen shows exactly what access the assistant is asking for. Depending on what it asks for, you will see one or more of these.

• Read your trust signal: your trust rankings, most trusted connections, bridges, and mirror scores. Text you submit for a mirror score is sent to Anthropic, a third-party sub-processor, to compute it.
• Suggest people you should meet, only if requested: the assistant surfaces people in and next to your trust network. This one only reads.
• Manage your events and invitations, only if requested: the assistant can show the events you host or are invited to, create, edit and delete events, suggest invitees, and send invitations and messages to guests that you confirm.
• Import your AI-drafted introduction and post your questions, only if requested. This one can write: if you ask your assistant to submit the introduction it drafted about you, it can propose profile updates for you to approve in the app. Up to five of its open questions get posted under your name, labeled "imported", and the first is sent to people who may be able to help. You can retract any of them at any time. People it names become suggestions for you to confirm.
• Post for you, only if requested. This one can write: the assistant can post under your name, exactly as if you had typed it. That means posting on your feed, where anyone can read it, and posting messages, thread replies and emoji reactions in your group channels, with images or text files attached. It can edit your posts and messages, and delete your channel messages. Deleting a message deletes everyone's replies under it, as in the app. It cannot read channel messages unless you also allow it to read your channels. New ways to post that UpTrust adds later come with this permission. You can delete any post in the app, a channel message while its channel is open.
• Read your groups and channels, only if requested: the assistant can list your groups and channels with how many unread messages each has, and read and search their messages, thread replies and who reacted, as you see them in the app. That includes private channels you are in and every private channel of a group you admin. It can see who is in your groups, including private ones and which members are admins, and which of your groups a given person is in. That is how it sees the answers to something it posted.

If the assistant asks for a new kind of access later, such as events, you will see the approval screen again. The one exception is Post for you: once you allow it, new ways to post messages are added without asking again.

What your assistant can do once connected

• Sort anything by your trust: "rank these five newsletters by how much my network trusts them."
• Check a source: "does my network trust this account?" The assistant gets a simple band (below, low, medium, high, or very high), never an exact score.
• Find your trusted voices on a topic.
• Show you bridges: things trusted on both sides of a disagreement in your network.
• Suggest people you might want to meet, through chains of trust.
• Plan an event: "invite people to golf this Saturday." The assistant fills in the details with you, creates the event, suggests invitees your network trusts, and sends the invites you approve.

Each tool, what it answers and the exact description your assistant reads are listed on UpTrust tools for AI assistants.

Here is what that looks like in a conversation.

Your privacy

• Connected assistants never see your raw trust scores. For trust questions they get orderings and coarse bands only. They also never get other members' email addresses.
• Text you ask an assistant to mirror-score is sent to a third-party service provider to compute the score.
• There are daily limits on how much a connected assistant can ask for. Normal use will not hit them.

Disconnect anytime

Open Settings and find the Connected Agents card. Click Revoke next to the assistant you want to remove, then confirm. Revoking cuts off access right away, and you can always reconnect later. For more detail, see connected AI agents.