---
title: "Gazebo — Credential Management and IAM for AI Agents"
description: "Secure AI agent credentials with scoped access, approval controls, audit logs, and instant revocation—without hardcoding or rotating shared API keys."
url: "https://gazebohq.com/"
---

# Gazebo — Credential Management and IAM for AI Agents

Gazebo is IAM (Identity and Access Management) for AI agents. AI agents need to call third-party APIs (Stripe, GitHub, Vercel, Cloudflare, OpenAI, etc.), but giving each agent a raw API key is unsafe. A full-access Stripe key handed to a Cursor agent gives that agent the ability to read customer data, modify subscriptions, delete payment methods, and configure webhooks — everything the key permits, with no scope boundary, no audit trail, and no revocation path that doesn't require rotating the key for every other system depending on it.

Gazebo applies the same principle as human IAM to AI agents:

- **Least privilege** — each agent gets access to exactly what its current task requires, nothing more
- **Full visibility** — every credential retrieval is logged: which agent, which service, timestamp
- **Instant revocation** — cutting one agent's access means deleting its profile, not rotating a shared key

## How it works

1. Connect your services to Gazebo once — Stripe, GitHub, Vercel, Cloudflare, OpenAI, Anthropic, and more. Credentials are encrypted with AES-256-GCM at rest. Plaintext values are never returned to the browser.
2. Create a named access profile for each agent. A profile defines exactly which services that agent can reach. The agent holds a profile token; the underlying API keys stay in the vault.
3. The agent authenticates to Gazebo's MCP endpoint with its bearer token. When it calls `get_credential`, Gazebo checks the profile policy, returns a scoped credential or denies the request, and logs the call.
4. To revoke an agent's access: delete the profile. The underlying service key is untouched; every other agent keeps working.