Let Claude Code into prod.

You decide what Claude Code can see and do. Hoop sits between the model and your infrastructure: it masks sensitive data, blocks destructive commands and logs every action.

Get started freeTalk to the team

Live on Product Hunt. Support the launch

claude code

Check why john@acme.com can’t log in to production

I’ll investigate by querying the production database through Hoop, starting with a read-only profile.

hoop exec db-prod-read
SELECT status, last_login, failed_attempts
FROM users WHERE email = 'john@acme.com'

status           disabled
last_login       2026-01-15 09:41:00
failed_attempts  12
email            j***@acme.com

The account is disabled after 12 failed login attempts on Jan 15. Hoop’s DLP policy masked the email. Want me to re-enable it?

Yes, re-enable the account

hoop exec db-prod-write
UPDATE users SET status = 'active',
failed_attempts = 0
WHERE email = 'john@acme.com'

Waiting for approval. Routed to DBA on-call.

#dba-approvals

hoop just now

Approval request

Requester
@alex via Claude Code
Connection
db-prod-write
Command
UPDATE users SET status='active'
AI risk
Low, 1 row affected

Claude asks. Hoop masks it.

Pick what you ask Claude Code. Every action it sends goes through Hoop first.

claude code → hoop gateway

> Show me the user with id 42
* Querying users table...
Found user 42:
  name:  "*********"
  email: "****@****.co"
  ssn:   "***-**-****"
3 values masked by hoop

PII, PCI and PHI invisible to Claude.

Claude gets full query context with names, emails, SSNs and card numbers redacted at the protocol layer. No schema configuration. No regex rules. The model never sees the raw data.

Minutes, not quarters.

  1. Connect

    Connect Claude Code to Hoop

    Point Claude Code at the Hoop gateway instead of straight at your infrastructure. One environment variable. No code changes.

  2. Define policies

    Define your policies

    Mask PII before Claude sees it. Block destructive commands. Route risky operations to a person for approval. Rules apply instantly.

  3. Ship

    Ship with confidence

    Claude Code works at full speed with full context. Your data stays protected. Hoop logs every session and can replay it.

From liability to advantage.

Six things change when Claude Code runs through Hoop.

  • Zero-config maskingPII, PHI and PCI redacted at the protocol layer before they reach the model context.
  • Destructive command blockingDROP TABLE, rm -rf, DELETE without WHERE. Blocked deterministically.
  • Human in the loopSchema changes and risky operations go to Slack for one-click approval.
  • Full session recordingEvery command, every response, every action logged and replayable for audit.
  • No credentials in the modelClaude calls internal systems through the gateway. Access is controlled at the protocol layer.
  • Deploys in minutesOne gateway. No agents, no plugins, no changes to Claude Code or your infrastructure.

Agents in prod are the risk now.

We built hoop.dev so your team controls what happens before something goes wrong.

Get started freeTalk to the team

Ask AI about Claude Code and hoop.dev