For the complete documentation index, see llms.txt. This page is also available as Markdown.

Apono Agent Privilege Guard for GitHub

Configure governed, just-in-time AI access to GitHub resources

Set up Apono Agent Privilege Guard for GitHub to give AI agents controlled, temporary access to GitHub resources.

An Apono admin configures the GitHub integration, an agentic access flow, and the GitHub MCP. Each user then connects a supported AI client through the Local Agentic Gateway, which brokers requests between the AI client, Apono, and GitHub.


Set up the Apono account

An Apono admin completes this setup once.

After the Apono account setup is complete, each user can connect a supported AI client to the Agentic Gateway.

Prerequisites

Requirement
Details

GitHub MCP integration

Connected Apono integration

Verify that the integration exists and is healthy on the Connected tab. If no GitHub MCP integration is connected, follow the GitHub integration docs.

Create an agentic access flow

Create a Self Serve access flow that allows agents to request temporary access to a GitHub instance on behalf of permitted users:

  1. Follow the procedure in Self Serve Access Flows to define the requestors, resources, access duration, and approval process. As you create the flow, apply the following settings for agentic access.

Setting
Configuration

Access Flow Name

Enter a unique, user-friendly name, such as Agentic - GitHub.

When Requestor

Select the users or groups on whose behalf an agent can request access. The agent can request access only on behalf of its associated user, so that user must match these requester conditions. For Can request for, select Themselves.

Request access to

Select your GitHub integration, and then select the resource types, resources, and permissions the agent can access.

Follow the Integrations instructions in the Define the resource section of the Self Serve Access Flows guide.

Grant for

Set the access duration to 5 minutes.

Approval

Select Automatic

Settings

Click the Restrict to agentic access toggle to enable the setting.

Allow Extend Duration does not apply to agentic access flows. When access expires, the agent requests access again.

  1. Configure any other Self Serve access flow settings according to your organization’s access policies.

  2. Click Save Access Flow.

Enable MCP features and the GitHub MCP

Follow these steps:

  1. On the Account Settings page, under AI & Automation, click the Enable MCP features toggle to enable the setting. The toggle turns green.

  2. On the MCPs tab, under Apono Managed Tools, click the toggle on the GitHub MCP card to enable the tool.

After setup, use the AI Agents tab to view discovered agents and their associated users. Use the AI Sessions tab to review agent activity, including access grants, tool calls, and intent policy decisions.


Set up the Agentic Gateway on a local machine

Learn more about local and remote agentic gateways setups.

Each user running the gateway completes this setup on their local machine.

Prerequisites

Requirement
Details

Apono account setup completed

Enablement of MCP feature for the Apono account

Confirm that an Apono admin has set up the Apono Account:

  • Connected the applicable integration

  • Created an agentic access flow with Restrict to agentic access enabled

  • Enabled MCP features and the GitHub MCP

macOS

Supported operating system

The Local Agentic Gateway is currently supported on macOS.

Node.js 20+

Required runtime

The gateway is launched with npx, which runs @apono-io/apono-mcp@latest, downloads the gateway binary, and manages the Apono sign-in. No separate Apono CLI installation is required.

Configure your AI client

If the Agentic Gateway is already configured for this client, skip to Verify your agent.

Follow these steps:

  1. Run the following command. If this is the first run, a separate browser tab opens.

  1. (First run only) Log in to Apono in the open browser tab.

Follow these steps:

  1. In Claude, from the main navigation, click Claude > Settings > Developer. The Settings page appears.

  2. Click Edit Config. The claude_desktop_config.json file opens.

  3. Add the following configuration.

  1. Save the file.

  2. Quit and restart Claude desktop.

  3. (First run only) Log in to Apono in the browser tab that opens.

Follow these steps:

  1. In Cursor, from the main navigation, click Tools & MCPs. The Tools page appears.

  2. Under Home MCP Servers, click New MCP Server. The mcp.json file opens.

  3. Add the following configuration.

  1. Save the file.

  2. Quit and restart Cursor.

  3. (First run only) Log in to Apono in the browser tab that opens.

Verify your agent

Perform the following tests:

  1. Ask the assistant to list your available integrations. The GitHub target should appear.

  2. Ask it to perform an action, such as "List the open pull requests in repository X". The agent will request access through the agentic access flow.

The first time a user accesses the GitHub MCP, the AI client prompts them to sign in with GitHub and grant OAuth consent. The client remembers this authorization, and all subsequent actions are performed as the authenticated user.

If either task fails, troubleshoot your setup.


Troubleshooting

Apono and GitHub Accounts

GitHub OAuth authorization fails.

Verify that the GitHub Client ID and GitHub OAuth Scopes are configured correctly, and that the OAuth App's Authorization callback URL matches the callback URL provided by Apono (GitHub OAuth Apps support only one callback URL).

After this, use any GitHub tool to trigger the AI client's sign-in prompt, then complete the OAuth consent in your browser. If authorization still fails, confirm with your administrator that both the GitHub (MCP OAuth) integration and the GitHub MCP tool are enabled.

Access is granted, but GitHub grants fail.

The configured GitHub OAuth Scopes may not permit this action. For example, repository write operations require the repo scope. Ask your administrator to review the configured scopes.

Local Machine

No targets or integrations are listed.

Follow these steps:

  1. Reauthenticate to Apono.

  1. Fully restart the AI client so the updated credentials take effect.

  2. If the target is still unavailable, confirm with an Apono admin that the platform setup is complete. This includes the integration, agentic access flow, MCP feature enablement, and applicable Apono Managed MCP.

  3. Follow the troubleshooting instructions in the applicable platform setup guide.

More details are needed on failures.

Add --debug to the gateway's args in your client config to enable verbose request/response logging, then check the gateway log.

Configuration changes are not reflected in the agent.

Fully restart the client. MCP servers are only launched at startup.

Last updated

Was this helpful?