Manage Agent Authority from an AI agent

Clicking through screens to answer one question, tag twenty servers, or draft a rule takes time you could spend on the decision itself. Agent Authority is also an MCP server, so you can do that work from the AI tool you already use, in plain language. Your agent calls Agent Authority with your own roles, your policies decide each call, and the audit log records every change.

What you can askCopy Anchor Link

Each request below is something an administrator would otherwise do screen by screen. The agent picks the tools, shows you what it found or plans to change, and runs only what your roles allow.

Answer questions and build your own reportsCopy Anchor Link

The read tools cover agents, MCP servers, policies, Agent Trail, analytics and findings, so the agent can answer questions the console splits across several screens:

• "Which tools did agents get blocked on most this week, and which rule blocked them?"
• "Which five people spent the most on models last month, and on which models?"
• "Which MCP servers did agents call most this month, and for whom?"
• "Show the open findings, most severe first, with the agent each one is about."

Ask for the answer as a table or a chart, and the agent builds it in your AI tool. Ask again next week and you get the same report on fresh data, a dashboard of your own without waiting for a screen to be built.

Tag resources in bulkCopy Anchor Link

Labels let one rule cover many servers or tools, but applying them one at a time is slow. Describe the set instead:

• "Create a finance label and apply it to Stripe, Ramp and Salesforce."
• "Label every MCP server that can send email outbound-comms."

The agent creates the label with tags_create and applies it with tag_assignments_set. Your next rule can then match the label instead of naming each server.

Draft and check policiesCopy Anchor Link

Describe the outcome, and let the agent write the rule:

• "Deny refunds in Stripe for everyone outside the finance group."
• "Let Claude read Jira but never change an issue."
• "Which tools can the agent called support-bot call right now?"

Ask the agent to read your current rules with policies_list before it adds one with policies_create, so it can tell you where the new rule lands in rule order and what it overrides. Review the rule it proposes before you let it write.

Run access reviewsCopy Anchor Link

• "Who holds the admin role, and when did each of them last make a change?"
• "List the API keys in the organization, and revoke the ones we no longer use."

How it stays safeCopy Anchor Link

• It acts as you. Each call runs with the roles of the person the agent belongs to. A viewer gets read tools only, and an admin also gets write tools. See Roles.
• Your policies apply. These tools go through the gateway like any other MCP server, so your rules, Data Protection filters and Usage Limits decide each call.
• Every change is recorded. Agent Trail records what the agent changed and the person it acted for.

Write tools can change Agent Authority itself, including the agent's own policies, API keys and role assignments. Deny the write tools an agent doesn't need, starting from the policy examples.

Turn it onCopy Anchor Link

1. Install the SecureAuth Agent Authority server from the MCP server catalog. It needs no app registration and no extra sign-in. See SecureAuth Agent Authority for the steps and the full tool list.
2. Connect your AI tool to the gateway if you haven't already, such as Claude, Claude Code, Cursor or ChatGPT.
3. Ask the agent "List my agents" to check that it reaches Agent Authority.

Next stepsCopy Anchor Link

You can now ask your AI tool to query Agent Authority, tag resources and draft rules, with your roles as the limit. Next, decide which write tools each agent should have.

• SecureAuth Agent Authority – setup, the tool list, and policy examples
• Roles – what each role lets the agent do for you