Gazebo is IAM for AI agents. Connect your API credentials once, create scoped access profiles for each agent, and let agents retrieve credentials at runtime — with every access logged and revocable in one click.

What Gazebo does

Instead of pasting API keys into agent configs, your agents call Gazebo at runtime to fetch the credential they need. Gazebo checks whether that agent is allowed to access that service, logs the request, and returns the credential — or denies it with a structured error.

The key properties:

• Scoped — each agent gets access to exactly the services it needs, with only the HTTP methods it requires
• Audited — every credential retrieval is logged, regardless of outcome
• Revocable — remove an agent's access instantly without rotating the underlying key

How to access Gazebo

There are two ways to connect an agent to Gazebo:

• MCP — for Cursor, Claude Code, Windsurf, and other MCP-compatible tools. Add Gazebo as an MCP server and use the get_credential tool.
• REST API — for custom agents, scripts, and any HTTP client. Same bearer token, same policy enforcement.

Where to start

• How Gazebo works — the full model in three minutes
• Quickstart — connect a service and retrieve a credential in five minutes
• MCP setup for Cursor — most common starting point

Open protocol

Gazebo is the reference implementation of the Agent Identity Protocol — an open standard for agent identity declaration, scoped permissions, and consent tokens. The protocol is not proprietary. Other tools and marketplaces can implement it independently.