Agent Keeper
Docs

Team Deployment

Roll out Claude Code security monitoring across your entire engineering team. One API key, one deployment path, full visibility.

How it works

Agent Keeper installs 4 lightweight hooks into Claude Code's lifecycle. Every prompt, tool call, and file operation passes through these hooks before it executes. The hooks run 55+ detection patterns on each evaluation, credential exfiltration, reverse shells, prompt injection, and more. Threats are flagged or blocked in real time.

Connect the hooks to your Agent Keeper dashboard and your security team gets fleet-wide visibility: every session, every tool call, every flagged event across all developers. No code changes, no SDK integration, no agent reconfiguration.

Deploy to your fleet

Three paths depending on your infrastructure. All use the same API key from Settings > API Keys in the dashboard.

MDM / JAMF (recommended for large teams)

Push a ~/.claude/settings.json config to every Mac via JAMF, Intune, or Kandji. Developers don't have to do anything, hooks activate the next time they start Claude Code.

Full walkthrough: JAMF / MDM Deployment

Repo hooks (recommended for specific repos)

Commit a .claude/settings.json to your repo. Every developer who clones it gets automatic coverage for that project. Good when you want per-repo policies rather than machine-wide coverage.

Setup: Claude Code Setup → Repo Hooks

Plugin (per-developer)

Each developer installs the Agent Keeper plugin and connects their account. Run these in Claude Code:

/plugin marketplace add agentkeeper/security
/plugin install agentkeeper
/agentkeeper:connect

Best for small teams or individual developers who want to get started fast.

Full reference: Plugin Reference

What your team gets

Security team / IT

  • Dashboard: every developer workstation, session, and tool call in one place.
  • Threat feed: real-time alerts for credential exfil, reverse shells, prompt injection, and 55+ patterns.
  • Audit trail: full history of every Claude Code action across the org.
  • Policies: enforce sandbox mode, block tools, restrict write paths, control which repos Claude runs in.

Developers

  • Zero disruption: hooks are fail-open with a 10-second timeout. If Agent Keeper is unreachable, Claude Code works normally.
  • No workflow changes: no new tools, no approval gates, no extra steps. Developers use Claude Code exactly as before.
  • Transparency: developers can run /agentkeeper:status to see their shield status and connection mode.

What data leaves the machine

Hook calls send: tool name, tool input summary, hostname, and session metadata. File contents are not sent. only file paths for Read/Write operations. Calls go outbound over HTTPS to your Agent Keeper app origin.

Set org-wide policies

Once workstations are connected, configure policies in Settings → Policies. Changes take effect within 30 seconds across all connected workstations.

  • Security level: strict, moderate, minimal, or paranoid.
  • Auto block: block threats vs. warn-only, defaulting to warn.
  • Blocked tools: prevent specific tools such as WebFetch or WebSearch.
  • Blocked commands: block bash commands containing specific strings.
  • Blocked write paths: prevent writes to sensitive paths such as /etc or ~/.ssh.
  • Require sandbox: enforce sandbox mode for all sessions.
  • Allowed git remotes: restrict which repos Claude Code can operate in.

For teams using an Identity Provider (Entra ID, Okta, Google Workspace), you can set per-group policy overrides, e.g., stricter rules for contractors.

Full reference: Team Policies · Identity Providers

Next steps

  1. Get your API key from the dashboard
  2. Pick a deployment path above
  3. Configure policies for your org
  4. Start a Claude Code session. Your first workstation should appear in the dashboard within seconds