Gazebo
    ServicesAgentsVaultsDocsWritingPricing
    Log inSign up
    Log in

    Getting Started

    • Introduction
    • How Gazebo works
    • Core concepts
    • Quickstart

    Agent Access

    • Access profiles
    • Bearer tokens
    • Audit logs

    MCP

    • Overview
    • Cursor
    • Claude Desktop
    • Claude Code
    • ChatGPT / Codex
    • Devin / Windsurf
    • Any MCP client
    • Get credential reference

    Authorization

    • Agent Consent Flow

    REST API

    • Overview
    • Authentication
    • Services
    • Agents
    • Access policies
    • Audit events
    • Consent administration
    • Agent credential access
    • Errors and limits

    Developer Tools

    • TypeScript SDK
    • CLI

    Security

    • Overview
    • Encryption model
    • Credential storage
    • Audit and compliance
    • Responsible disclosure
    Documentation

    Getting Started

    • Introduction
    • How Gazebo works
    • Core concepts
    • Quickstart

    Agent Access

    • Access profiles
    • Bearer tokens
    • Audit logs

    MCP

    • Overview
    • Cursor
    • Claude Desktop
    • Claude Code
    • ChatGPT / Codex
    • Devin / Windsurf
    • Any MCP client
    • Get credential reference

    Authorization

    • Agent Consent Flow

    REST API

    • Overview
    • Authentication
    • Services
    • Agents
    • Access policies
    • Audit events
    • Consent administration
    • Agent credential access
    • Errors and limits

    Developer Tools

    • TypeScript SDK
    • CLI

    Security

    • Overview
    • Encryption model
    • Credential storage
    • Audit and compliance
    • Responsible disclosure
    DocsMCPClaude Code

    Claude Code

    Documentation index (llms.txt)View as Markdown

    Add Gazebo to Claude Code as an MCP server so it can retrieve credentials from your vault during coding sessions.

    Prerequisites

    • A Gazebo account with at least one service connected
    • For the desktop connector: Claude Code access and permission to approve selected keys in Gazebo
    • For terminal configuration: an agent created in Gazebo with a bearer token

    Desktop connector

    In Claude's Code tab, start a local session in a separate test folder. Open + → Connectors, then add or reconnect Gazebo using https://app.gazebohq.com/api/mcp. Complete Gazebo's OAuth sign-in and approve only the service keys this connection needs. An existing Gazebo connector may already appear in the menu.

    Start with get_identity. OAuth credential retrieval must include the approved key_name. Use a fictional key for the first retrieval, check its event with list_audit_events, then revoke the grant in Gazebo and confirm a fresh request fails without reconnecting.

    The Code tab and normal Chat mode are different test surfaces, even when they share an account connector. Selecting This computer does not make the remote connector local-only: Anthropic's connector infrastructure can receive returned credentials.

    Terminal configuration

    In Claude Code's MCP server configuration, add Gazebo with endpoint https://app.gazebohq.com/api/mcp and the header Authorization: Bearer <your-agent-token>. Create the agent bearer token at Gazebo Agents.

    For a project .mcp.json, the current documented HTTP configuration is:

    {
      "mcpServers": {
        "gazebo": {
          "type": "http",
          "url": "https://app.gazebohq.com/api/mcp",
          "headers": {
            "Authorization": "Bearer ${GAZEBO_AGENT_TOKEN}"
          }
        }
      }
    }
    

    Set GAZEBO_AGENT_TOKEN in the environment inherited by Claude Code. Do not put the real token in a committed file. Note that Claude Code uses ${NAME} interpolation, unlike Cursor's ${env:NAME}. See Claude Code's current MCP documentation. Configuration was checked against that documentation; it is not a completed runtime test of your installed release.

    Use the current Claude Code documentation for terminal configuration rather than assuming that the desktop connector's setup applies to the CLI.

    Tested status

    Claude Code has a tested desktop OAuth path. Identity, named fictional-key retrieval, matching audit retrieval and fresh-request rejection after revocation passed on 6 November 2026.

    See the compatibility guide for the macOS test environment and evidence limits. Terminal/CLI tool execution, token renewal, timed expiry and marketplace-plugin installation remain unverified; the earlier CLI check covered connection health only.

    Verify the connection

    Ask Claude to call get_identity:

    get_identity()
    → {
        status: "ok",
        name: "claude-dev",
        accessible_services: [
          { service: "github", allowed_methods: ["GET"] },
          { service: "stripe", allowed_methods: ["GET", "POST"] }
        ]
      }
    

    Retrieve a credential

    get_credential({ service: "github", key_name: "token", method: "GET" })
    → "<actual-stored-provider-key>"
    

    Claude Code receives the actual provider credential. The declared method controls retrieval; the provider key's permissions control subsequent GitHub API calls. Do not persist the returned value. Revoking Gazebo access blocks new retrieval, not use of a key already received.

    Usage guidance for Claude

    Claude works well with Gazebo when given context about what it's allowed to do. You can prompt it directly:

    "Use the Gazebo MCP server to get my GitHub token, then list the open issues on my repo."

    Claude will call get_identity to confirm access, then get_credential to retrieve the token, then use it for the GitHub API call.

    Troubleshooting

    MCP server not appearing — Confirm the Claude Code MCP configuration follows Claude Code's current documentation and that the endpoint and header are correct.

    401 Unauthorized — Token is wrong or the agent was deleted. Check the Gazebo dashboard.

    Denial on get_credential — The method you requested isn't in the agent's access profile. Update the profile in the Gazebo dashboard.

    Explore Gazebo

    Gazebo gives each agent an independent access boundary around your agent credentials, with controls for approval, auditing, and revocation.

    Scoped identities for AI agentsGive every agent only the credentials it needs.Credential security and audit controlsSee how access is protected, recorded, and revoked.
    Claude DesktopChatGPT / Codex
    Gazebo

    Credential management for AI agents. Control credential retrieval, review access logs, and revoke an agent’s future retrieval without rotating shared API keys.

    Gazebo

    • Account
    • Demo
    • Pricing

    Explore

    • Services
    • Agents
    • Vaults

    Developers

    • Docs
    • SDK
    • CLI
    • API
    • MCP

    Content

    • Writing
    • Topics
    • Blog

    Company

    • About
    • Contact
    • hello@gazebohq.com

    © 2026 Gazebo. All rights reserved.

    PrivacyTermsSecurity