unbound
Use Claude Code without fear of data leakage. Unbound enforces command policies, DLP and jailbreak guardrails, and streams session analytics
Open source Open in the app JSON README (API)
About
Use Claude Code without fear of data leakage. Unbound enforces command policies, DLP and jailbreak guardrails, and streams session analytics to your security dashboard — all with zero-latency fail-open design.
Details
- Kind
- Plugins
- Topic
- Marketing & analytics
- Publisher
- websentry-ai
- Origin
- marketplace
- Category
- ferramentas
- Last push
- 2026-05-05T16:13:46Z
- Repository state
- ativo
- Language
- Python
- License
- MIT
- Added
- 2026-08-30 01:48:58
- Updated
- 2026-08-30 01:48:58
- Origin id
websentry-ai/claude-code-plugin/unbound
README
# Unbound — Security & Governance for Claude Code Real-time policy enforcement, DLP guardrails, and session analytics for [Claude Code](https://claude.ai/code) — powered by [Unbound AI](https://getunbound.ai). ## Why Unbound? Engineering teams adopting Claude Code face real risks: sensitive data pasted into prompts, dangerous commands executed without oversight, and zero visibility into what's happening across sessions. Unbound solves this by sitting between your developers and Claude Code — enforcing security policies in real time and streaming every session to a central dashboard. No workflow disruption, no developer friction. ## Quick Start ### Step 1 — Install the plugin Inside Claude Code, run these two commands: ``` /plugin marketplace add websentry-ai/claude-code-plugin /plugin install unbound@websentry-ai/claude-code-plugin ``` ### Step 2 — Run setup Open Claude Code and run: ``` /unbound:setup ``` The setup skill walks you through getting an API key, persisting it to your shell profile, and verifying connectivity. Takes under 5 minutes. ### Step 3 — Verify on your dashboard After setup, every tool invocation is checked against your Unbound policies. Test with: - **Command policy**: create a BLOCK rule in your Unbound dashboard, then try running the blocked command — it should be rejected - **DLP guardrail**: enable DLP, then type a prompt containing an SSN — it should be blocked - **Analytics**: run any command, then check your Unbound dashboard for the event ## What Gets Enforced | Capability | What it does | |---|---| | **Command policies** | Block or warn on dangerous tool invocations before they execute | | **Prompt guardrails** | DLP, NSFW, and jailbreak detection on every user prompt | | **Audit logging** | Streams tool usage to the Unbound dashboard in real time | | **Session analytics** | Sends the full conversation exchange when a session ends | ## Fail-Open Design The plugin never blocks your workflow if the Unbound API is unreachable or the key is missing. All hooks fail open — Claude Code continues working normally, and queued events are replayed when connectivity resumes. ## Enterprise Deployment For fleet deployment where end users cannot disable the plugin, see the [enterprise deployment guide](enterprise/README.md). In short: drop a managed settings file via your MDM and provision `UNBOUND_CLAUDE_API_KEY` per device. ## Configuration | Variable | Description | |---|---| | `UNBOUND_CLAUDE_API_KEY` | Bearer token for the Unbound API. Get one at [app.getunbound.ai](https://app.getunbound.ai) → Settings → API Keys | ## Logs | Path | Contents | |---|---| | `~/.unbound/logs/debug.jsonl` | Raw stdin from every hook event (for debugging) | | `~/.claude/hooks/agent-audit.log` | Per-session audit trail | | `~/.unbound/logs/offline-events.jsonl` | Events that failed to send (replayed on reconnect) | | `~/.claude/hooks/error.log` | API errors (last 25 entries) | ## Troubleshooting **Plugin not loading** Run `claude plugin list` and confirm `unbound` appears. If not, re-install with `/plugin marketplace add websentry-ai/claude-code-plugin` followed by `/plugin install unbound@websentry-ai/claude-code-plugin`. **API key not set** Run `/unbound:setup` again — it will detect the missing key and guide you through setup. You can also set `UNBOUND_CLAUDE_API_KEY` manually in your shell profile. **Command blocked unexpectedly** Check your policy rules at [app.getunbound.ai](https://app.getunbound.ai). The block response includes a reason — review it to confirm the rule that triggered. **Events not appearing on dashboard** Verify your API key is valid and the Unbound API is reachable. Check `~/.claude/hooks/error.log` for details. Offline events are stored in `~/.unbound/logs/offline-events.jsonl` and replayed automatically. ## Support Questions or issues? Email [support@unboundsecurity.ai](mailto:support@unboundsecurity.ai). ## License [MIT](LICENSE)