Monitor prompts, tool calls, file access, and MCP activity on every workstation where the plugin is installed. Apply policy before an action runs and keep a record of each decision.
Claude Code security means seeing and governing what the agent does on a developer's machine: the commands it runs, the files it reads and the MCP tools it calls. AgentKeeper by RAD Security adds hooks through a Claude Code plugin that record each action and check it against your policy in monitor, warn or block mode.
/plugin marketplace add rad-security/claude-code-plugin
/plugin install agentkeeper
/reload-plugins
/agentkeeper:connectApprove the device in your browser, then restart Claude Code to activate hooks.
No credit card required · Free tier forever · Works instantly with no account
Claude Code acts on its own: it edits files and runs commands without waiting for you. That is the surface to govern.
Claude Code has had critical vulnerabilities including RCE and API key exfiltration. Keeping up manually doesn’t scale.
Your developers run Claude Code on their laptops. You have no idea what commands it runs, what files it reads, or where it sends data.
Anthropic gives you managed settings. But there’s no centralized policy engine, no audit trail, and no fleet-wide compliance view.
Four layers of protection. One command to activate.
Install AgentKeeper from the Claude Code plugin marketplace. Hooks, threat detection, and 9 security skills activate instantly — no account, no config.
/plugin install agentkeeperRun /reload-plugins, connect, then restart Claude Code so hooks load at startup.
Or for team-wide push-hooks: agentkeeper claude-code init
Four hook types give you full visibility into every Claude Code session across your org.
UserPromptSubmitScans every prompt before Claude processes it
PreToolUseEvaluates tool calls before execution
PostToolUseMonitors tool output for hidden instructions
SessionStartValidates environment and policy on launch
Behavioral detection patterns tuned for zero false positives on normal development. Dangerous tool calls are stopped before execution.
One dashboard for your entire Claude Code fleet. Compliance percentage, active developers, blocked threats, version tracking.
94%
Compliance
23
Developers
147
Threats Blocked
312
Sessions / 24h
Every feature designed to give you control without slowing your developers down.
UserPromptSubmit hook scans every prompt before Claude processes it. Catches social engineering, jailbreaks, and embedded instructions in file content. All prompts logged to the threat feed.
Allow, warn, or block specific tools per org. Enforce which Bash commands, file operations, and web fetches your team can use.
Restrict Claude Code to approved repositories only. Prevent code from being pushed to unauthorized remotes or cloned from untrusted sources.
Require sandbox mode across your entire fleet. Verify every developer session runs in a restricted environment with no exceptions.
Every prompt and tool call logged to the threat feed with timestamps, session context, and user identity. Full forensic trail for incident response and compliance.
Claude Code version, sandbox status, hooks active per developer. One view for your entire fleet’s security posture.
Anthropic Enterprise is a great start. AgentKeeper fills the gaps.
| Capability | Anthropic Enterprise | + AgentKeeper |
|---|---|---|
| Managed settings | ||
| Permission rules | ||
| Usage analytics | ||
| Real-time threat blocking | ||
| Behavioral detection patterns | ||
| Prompt injection monitoring | ||
| Fleet compliance dashboard | ||
| Tool-level policy per org | ||
| Git remote validation | ||
| Complete audit trail |
Start free — no credit card required. Scale when you need to.
For individual developers
Need enterprise features? Contact us for enterprise pricing
Updated
It adds hooks for UserPromptSubmit, PreToolUse, PostToolUse and SessionStart. They inspect prompts and tool calls, match them against detections and your organization's policy, and record the decision with its session and workstation context.
Only when a policy is in Block mode. Monitor records silently, Warn tells the person and refuses nothing, and Block refuses the call. Hooks fail open: if AgentKeeper cannot evaluate a call, Claude Code carries on instead of stopping the developer.
Run /plugin marketplace add rad-security/claude-code-plugin, then /plugin install agentkeeper, /reload-plugins and /agentkeeper:connect. Approve the device in your browser and restart Claude Code. The Free plan covers one workstation and needs no credit card.
In local-only mode the plugin makes no network calls; detection runs on the workstation. After you connect an account, hooks call the AgentKeeper API and decision metadata for matched tool calls goes to your dashboard. The plugin is MIT-licensed, so you can read the source.
Managed settings, permission rules and the sandbox are Anthropic's native controls, and you should keep using them. AgentKeeper adds one policy and one record across Claude Code and the other coding agents your teams run, attributed to each person and workstation.
/plugin marketplace add rad-security/claude-code-plugin
/plugin install agentkeeper
/reload-plugins
/agentkeeper:connectApprove the device in your browser, then restart Claude Code to activate hooks. Reconnect later with /agentkeeper:connect
agentkeeper claude-code initPush-hooks commit to git — every developer who pulls gets covered automatically.
No credit card required · Setup in 30 seconds