SecureAuth Agent Authority
secureauth.com

Using SecureAuth Agent Authority

How to connect your AI assistant and the services you use, and see your activity through SecureAuth Agent Authority

A short guide for everyone in your organization who uses an AI assistant such as Claude Desktop or Claude Code to work with company tools and data. It's written for end users, not administrators: your IT team has set up SecureAuth Agent Authority and handles the configuration, so there's nothing for you to set up or understand. This page walks you through connecting your assistant, connecting the services you use, and seeing what your assistant has done on your behalf.

Your user portal link

Your administrator will give you a link to the user portal. It opens the Quickstart page, which lists the AI assistants your organization supports and the steps to connect each one. If you don't have the link yet, ask your administrator.

Administrators: open Agents & NHIs and click Copy portal link to get the link to share. See Share the portal link.

Why your organization uses Agent Authority

Your assistant can act on your behalf: draft a reply, post an update, open a pull request. Agent Authority is what lets your organization allow that safely. It gives your assistant an identity tied to you, so your assistant works with your own accounts and never reaches further than you can, and it keeps a record of what was done.

What that means day to day depends on the work you do:

What your assistant can do depends on the work you do. Three examples, each passing through SecureAuth Agent Authority, which applies your own limits: for everyday work in Microsoft 365 the assistant can read mail and draft replies but cannot send, delete, or forward it; for team updates in Slack it can read and post in one channel but cannot reach the rest of Slack; for working on code in GitHub it can open and review pull requests but cannot delete code or skip review. These are examples only, and your organization's administrators decide what your assistant can do.Agent AuthorityApplies your limitsEveryday workMicrosoft 365Can read mail and draft repliesCannot send, delete, or forward itTeam updatesSlackCan read and post in one channelCannot reach the rest of SlackWorking on codeGitHubCan open and review pull requestsCannot delete code or skip review

These are examples. Your administrators decide which tools your assistant can use and which actions it can take, so your own setup will differ.

What you authorize, and when

Agent Authority asks you to authorize two different things. They are separate steps, not alternatives.

What you authorizeHow oftenWhat it allows
Your AI assistantOnce per assistantThe assistant can act as you through Agent Authority.
Each service accountOnce per serviceAgent Authority can use your own Slack, Jira, or GitHub account on your behalf.

The two are independent: connecting your assistant does not reach any service on its own, and a connected service account sits unused until you connect an assistant too. Most people connect an assistant first, then connect each service the first time they need it.

Connect your AI assistant

Your IT team has already added the SecureAuth AI Gateway connector to your AI assistant. You don't need to install or configure anything. You just need to find the connector, click it, and authorize it to act on your behalf.

Authorizing opens a sign-in page in your browser. After you sign in, you're granting Agent Authority permission to interact with your organization's tools and resources on your behalf, always using your own identity.

Claude Desktop

  1. In Claude Desktop, go to Settings > Connectors.

  2. Find SecureAuth AI Gateway in your list of connectors and click it.

  3. Click Connect. A browser window opens for sign-in.

  4. Sign in and approve. Your assistant is connected. You're ready to go.

Claude Code

  1. Start Claude Code in your terminal:

    claude

  2. Type /mcp and select SecureAuth AI Gateway.

  3. Choose Authenticate. A browser window opens for sign-in.

  4. Sign in and approve. Your assistant is connected. You're ready to go.

Your assistant now appears on the Agents page.

Using a different assistant?

Cursor, Codex, GitHub Copilot, and more are also supported. Open the Quickstart page in the Agent Authority user portal for the current list and the exact steps for each one.

Connect your accounts

Agent Authority always acts using your accounts. You connect each service once, and you can do it either of two ways. Both give the same result, so use whichever you meet first.

Wait for your assistant to ask

The first time your assistant tries to use a tool that needs sign-in, for example sending a Slack message, it asks you to connect that account:

  1. Your assistant tries to use a tool that requires sign-in.
  2. Agent Authority replies with a link asking you to connect that account.
  3. You click the link and sign in with the service the normal way.
  4. Agent Authority securely stores the result and uses it for you from then on. You don't need to sign in again for that service.

Connect it yourself first

Open the Connections page to see every service available to you and connect one before your assistant needs it.

The Connections page in the user portal, listing each service with a Connected or Not connected status and a Connect or Disconnect button
Your own Connections page. Each service is either Connected or Not connected.

Each service shows one of two statuses:

  • Connected. Ready to use. Select Disconnect to remove the connection and delete the stored access.
  • Not connected. Select Connect and sign in to the service.

A working connection keeps working without you signing in again, because Agent Authority renews your access in the background. If the service provider withdraws access, for example because you removed the app in that service's own settings, the service goes back to Not connected and the next tool call gives you a fresh link.

Your connections are yours alone

An assistant acting for you can only use your connections and can only see what your account can see. Another person's assistant can never reach your accounts, and vice versa.

What you'll see

Open your user portal link and sign in. In the left menu you'll find a group called AI Gateway with five pages:

The user portal Dashboard, with the AI Gateway menu group listing Dashboard, Quickstart, Agents, Connections, and Activity
The portal as you see it. The AI Gateway group holds your five pages, and every figure on the page counts only your own activity.
  • Dashboard. Your own activity at a glance, explained in What your dashboard is telling you.
  • Quickstart. Connect an AI assistant.
  • Agents. The assistants you've connected. Open one to see its client name and version, when it was last active, and which tools it can use.
  • Connections. The service accounts you've connected, such as Slack or Jira.
  • Activity. A running list of what your assistants have done for you: which tool, when, and whether it succeeded. This is your own audit trail.

Everything here is yours only. You won't see anyone else's assistants or activity, and they won't see yours.

To stop an assistant from acting for you, disconnect it from Agents. To remove access to a service, use Connections.

What your dashboard is telling you

The Dashboard answers three questions about your own AI use: whether your assistants are actually working, where they keep getting stopped, and roughly what they cost. Every figure counts only your activity, over the date range you pick at the top right, starting at Last 7 days.

The cards along the top

CardWhat it meansWhy you might care
Active Agent InstancesAssistants of yours that did something in this range.Tells you which of your assistants you actually use.
Unique Tools UsedHow many different tools those assistants called.Shows how far your assistant reaches on your behalf.
Blocked Tool CallsCalls your organization's rules refused. Shown in red, but not a mistake you made.The one number worth acting on. A high count means your assistant keeps hitting a limit, so ask your IT team if you need that access.
CostAn estimate of what your usage would have cost at list price, which nobody paid.A rough sense of what your AI use is worth. It is not your organization's invoice.

The percentage under a figure compares your chosen range against the previous stretch of the same length. If you see No LLM usage yet in place of the cost card, your assistants have not reported any AI usage. Connect one from Quickstart.

Your token usage

Tokens are the units AI models count and bill by. Once your assistants report AI usage, a token section appears under the spend chart:

ElementWhat it shows
Tokens Over TimeYour tokens per day, with a line for the previous period.
Token breakdownYour total tokens, split into input, output, cache read, and cache creation. Cache read tokens are context the model reused, which costs less.
Top Models by TokensThe models you use most.

The charts below

ChartWhat it shows
Spend Over Time (USD)Your cost per day, with a shaded band for the part that is real metered usage and a dashed line for the previous period.
Tool Executions Over TimeHow many tool calls you made per day, gaining a red line once you have blocked calls.
Activity by Hour (UTC)The hours you work through your assistants. It reads in UTC, so it may not match your own clock.
Top Tools and Top Agent InstancesWhat you use most. Each has a blocked counterpart showing what gets refused most.

FAQ

Identity and sign-in

Add people to the workspace identity pool, or federate the directory you already run, and let their groups drive policy

Use cases

Task-focused walkthroughs: real policies, real agents, real resources — from read-only Jira access for Claude to per-team token spend.

On this page

Why your organization uses Agent AuthorityWhat you authorize, and whenConnect your AI assistantClaude DesktopClaude CodeConnect your accountsWait for your assistant to askConnect it yourself firstWhat you'll seeWhat your dashboard is telling youThe cards along the topYour token usageThe charts belowFAQ