Connecting an AI Agent to AwardLettr

AI Agent Access lets you connect an AI assistant, Claude, ChatGPT, or Codex, straight to your AwardLettr workspace, so you can ask it about your cases, have it draft time entries, or hand it an assignment email and let it stage the new file, all without leaving your chat. It only sees what you let it see. Anything that creates or reschedules a case waits for your approval, and a few small, easy-to-undo actions like notes and reminders happen right away.

Useful if you already talk to an AI assistant during your day and want it to actually know your workspace instead of you copying and pasting case details into it. It is also the backbone behind Jaxson, AwardLettr's built-in coach, which uses this same connection to interview you and build your agent playbook (see "Teaching Your Agent How You Work").

Two Ways to Connect

How you connect depends on the app. Sign-in apps, like Claude Desktop and ChatGPT, connect with your AwardLettr login, the same way you would sign into a website. Command-line tools, like Claude Code and Codex, do not have a browser to sign in with, so they use a key you create on this page instead.

App	How it connects	Do you need a key?
Claude Desktop	Sign in with your AwardLettr login and approve on a consent screen	No
ChatGPT	Sign in with your AwardLettr login and approve on a consent screen. On Business, Enterprise, and Edu plans a workspace owner or admin creates the app.	No
Claude Code	Paste a ready-made connect command	Yes
Codex	Paste a connect command and store the key in your environment	Yes

Either way, it acts as you

A connected agent, sign-in or key, acts as the person who connected it and only within the permissions granted. If that person loses access to the workspace, the connection stops working right away.

Creating a Key

Command-line tools like Claude Code and Codex need a key. To create one:

Go to Settings, then Integrations, then AI Agent Access

Owners and admins can reach this page. It is not visible to staff.

Click "Create key"

Give it a name that reminds you where it is used, for example "Claude on my laptop."

Choose what it can do

Check the boxes for what this key is allowed to read and do. See "Picking What It Can Do" below.

Copy your key

It is shown once. Copy it now, or use one of the ready-made connect snippets shown right next to it.

The key is shown once

For your security, AwardLettr does not store your key anywhere you can look it up again. If you lose it, revoke it and create a new one.

Picking What It Can Do

Every key carries its own list of permissions, checked one by one when you create it. Some let the agent read information, like your cases, your schedule, your documents, or your unbilled totals. Others let it draft or write something, like a note, a reminder, a time entry, or a new case file. Give a key only what you actually want it to do. You can create more than one key with different permissions for different tools, and revoke any of them independently at any time.

Connecting Your Client

The AI Agent Access page includes a "How to connect" guide with tabs for Claude Desktop, ChatGPT, Claude Code, and Codex, each with its own copy-paste steps and the same address for every one of them. If you just created a key, the show-once panel also offers a "let Claude set itself up" prompt: copy it, paste it into a new Claude Desktop chat, and Claude walks you through the setup step by step, no technical background needed.

Connecting ChatGPT

ChatGPT calls these custom MCP apps, and where the setting lives depends on your plan. On ChatGPT Business, Enterprise, and Edu, a workspace owner or admin has to turn on custom connectors for the workspace before anyone sees the option, and the app is created from Workspace settings. On Plus and Pro you turn on Developer mode in your own settings. Either way, do it at chatgpt.com in a browser, since custom apps are web only and the phone app cannot create or use them.

Turn on custom connectors for the workspace (Business, Enterprise, Edu)

A workspace owner or admin opens Workspace settings, then Permissions & roles, then Connected data, and turns on Developer mode (also labeled "Create custom MCP connectors"). On Business, only owners and admins can use Developer mode. On Plus and Pro, skip to the next step.

Turn on Developer mode for yourself

Open Settings, then Apps, then Advanced settings, and switch on Developer mode. ChatGPT may also offer to turn it on when you click Create.

Create the app

Business admins: Workspace settings, then Apps, then Create. Everyone else: Settings, then Apps, then Create. Name it AwardLettr, paste the address from the AI Agent Access page as the MCP server URL, choose OAuth for authentication, and leave any client ID or client secret fields blank.

Scan tools and sign in

Click Scan Tools. Sign in to AwardLettr in the window that opens, click Authorize on the consent screen, wait for the scan to finish, then click Create.

Use it, and publish if you want your team on it

The app works for you right away as a draft. In a new chat, click +, pick AwardLettr from the apps list (or type @AwardLettr), and ask "call whoami on AwardLettr." To let teammates use it, open Workspace settings, then Apps, then Drafts, and click Publish.

What each ChatGPT plan can do

Business, Enterprise, and Edu get full read and write. Plus and Pro can connect but only read, so creating a case or logging time from ChatGPT needs a workspace plan. The Free plan cannot add custom connectors at all. Those limits come from OpenAI, not AwardLettr, and Claude reads and writes on any plan. One more Business quirk: once an app is published, ChatGPT freezes what it can do, so when AwardLettr adds new abilities, recreate and republish the app to pick them up.

What Your Agent Can Do

Within the permissions you grant, a connected agent can read case details, your schedule, notes, documents, time entries, and unbilled totals. It cannot read your directory of carriers, adjusters, appraisers, and umpires, though as covered below it can add a new contact to it.

For anything that changes your workspace, almost everything is propose first, you approve second. Your agent can draft a new appraisal for you to review before it is created, propose a change to a date on an existing case (today that covers the inspection date and a tentative inspection date), draft a time entry you confirm on the appraisal's Financials tab, or draft new house rules for your playbook. It can also add an internal case note, create a reminder, and add a new directory contact directly since those are easy to undo and never notify anyone outside your workspace. Uploading a document is the one exception that lands right away rather than as a proposal, since there is nothing to approve about adding a file.

Nothing gets created without you

A new appraisal or a change to an existing one only happens after you review it and click approve. Your agent cannot create a case, change a date, or send anything to anyone on its own.

Reviewing What Your Agent Proposes

Proposed new case files and proposed changes to existing cases show up on My Work, under "New case files from your agent" and "Proposed changes from your agent." For a new case, click "Review & create" to open the regular New Appraisal form, already filled in, so you can check it over, fix anything, and save it the normal way. For a change to an existing case, click "Approve" to apply it, or "Dismiss" to ignore it. Draft time entries show up on the appraisal's Financials tab for you to confirm, the same place manual suggested entries already appear.

Where It Cannot Go

A connected agent cannot send an email, a booking invitation, or a message to anyone on your behalf. It cannot touch billing, your team, or your subscription, and it cannot delete anything. If it does not have a permission checked on its key, it simply cannot do that thing, no matter what it is asked.

Revoking a Key

Open Settings, then Integrations, then AI Agent Access, find the key in the list, and click "Revoke." Access stops immediately and cannot be undone, so create a new key if you need to reconnect that tool later.

Checking What It Has Done

Every call a connected agent makes is recorded, and the AI Agent Access page shows when each key was last used. You can also just ask your agent directly, for example "what have you done in AwardLettr recently," and it can look up its own recent activity for you.