Give a Team Governed Coding Agents
Who this is for: the Admin who puts a development team's coding agents behind HYPR AgentPass™, with each developer as the Supervisor of their own agent.
Purpose
This playbook gives the running order for putting a team's coding agents behind AgentPass. Each developer then runs Claude Code through the gateway, with the tools their role allows, and approves from a phone.
Each task is a summary. The detail, with screenshots, field values and caveats, is on the page the task links to.
When you finish:
- A developer's first agent connection is held for approval.
- The agent sees only the MCP tools its supervisor's roles permit.
- The Request Log attributes every request to a person.
Architecture
A local agent's traffic reaches the gateway, either because the developer points the agent at it, or because Endpoint Guard on the workstation routes it there. The gateway checks the agent's identity and its supervisor's roles, then forwards to the provider or an MCP connector. Approvals go to the supervisor's phone. For the request path, see Architecture: The Request Path.
Prerequisites
- Role: Admin. Connectors, role mappings and gateway settings need it.
- An OpenID Connect provider for console sign-in
- A HYPR tenant for supervisor passkeys
- An account with the AI provider the team uses
- The endpoints and credentials of the MCP servers the team needs
- To route workstations centrally, Endpoint Guard deployable to the team's computers
For the full list, see Prerequisites.
Procedure
Complete these tasks in order.
Map Roles, Then Turn On Console Sign-In
Configure console SSO against your identity provider, and save it without turning it on. Then, on the Roles tab of People, map your administrators' directory group to AgentPassAdmin, and any other groups to their roles, under Identity provider. Developers need no mapping: anyone whose values map to no role gets Supervisor. Finally, turn on Enable SSO, and sign in with SSO to confirm your own role is Admin. The Users tab of People shows who has which role.
See Signing In to the Console and Map Identity Provider Values to a Role.
Add the AI Provider Connector
Add the connector for the team's provider, and set its Onboarding Mode to Require Approval. Each new agent's first connection is then held until its supervisor approves it. On an Anthropic connector, the Connect AI tools card on the Overview tab shows the settings to give your developers.
See Connect Your Developers and Onboarding Mode.
Add and Publish the MCP Connectors
Add one MCP Server connector for each tool the team needs. Leave Publish to endpoint catalog on, so managed workstations pick the connectors up automatically. A developer who adds a connector to their agent by hand uses its Gateway MCP URL.
See MCP Servers and Provisioning MCP Tools.
Scope Tools by Role
In each role's Agent tools permission section, permit only the tools that role's agents may call. An agent is offered only the tools that one of its supervisor's roles can use. For most developers, that is the Supervisor role.
See Controlling MCP Tool Access by Role.
Route the Team's Workstations
Choose how each developer's agent reaches the gateway.
- With Endpoint Guard. On Desktops, create an enrollment token on the Installation tab, and deploy Endpoint Guard to the team's computers with it. Then, on Desktops → Desktop configuration → Configuration, turn on Configure agents from AgentPass, turn on Claude Code, keep Route model traffic through AgentPass ticked, and pick your connector. Enrolled computers pick up the configuration on their next check, so developers change nothing themselves. To keep the agent from reaching its provider directly, tick Block direct traffic.
- By hand. Each developer sets the agent's provider base URL to the gateway, such as
ANTHROPIC_BASE_URLfor Claude Code, from the settings in the Connect AI tools card.
See Agent Configuration and See and Govern the AI Tools Already on Your Workstations.
Set Up Mobile Approvals
Connect the gateway to your HYPR tenant, so supervisors can enroll a passkey. Then configure push delivery, so approval requests reach their phones. A test push confirms the path.
See Connecting to Your HYPR Tenant and Notifications.
Hand the Supervisor Guide to the Team
Start each developer at Being a Supervisor, and give them Using AI Tools with HYPR AgentPass for what they see in the tool. The supervisor guide walks them through these tasks in order:
- Install the HYPR app for iOS or Android. See Getting the HYPR App on iOS or Getting the HYPR App on Android.
- Enroll a passkey. See Setting Up Your Approval Passkey.
- Approve their agent's first connection. On a computer Endpoint Guard routes, the agent already points at the gateway, so they restart it and sign in as usual. Otherwise, they point it at the gateway first. See Onboarding Your Agent.
- Add the MCP connectors they need, unless Endpoint Guard provides them. See Connecting to a Gateway MCP Connector.
Validate
Have one developer start their agent through the gateway, and check the following:
- Their phone receives the onboarding request, and approving it releases the agent's first request.
- On Activity → Request log, that request is attributed to the agent and its supervisor.
- The agent's tool list shows only the role's permitted tools.
- On a routed computer, the By desktop tab of Activity → Desktop activity lists the computer, and its Managed configuration shows routing as Applied.
See Approving and Denying Agent Actions.
Logs and Audit Trail
Every request the team's agents send is in the Request Log. Agent session forensics reconstructs a whole coding session: the files touched, the commands run and the tools called.
See Request Log and Agent Session Forensics: Overview.
Related
- Local Agents — what Endpoint Guard does for each coding agent
- Require a Person for High-Value Agent Actions — add approval policies once agents are onboarded
- See and Govern the AI Tools Already on Your Workstations — enforce routing for tools the team already runs