Watchgoose MCP Server

Watchgoose MCP

Watchgoose MCP connects MCP-capable AI clients to the Watchgoose Management API. It lets you inspect cron and recurring-task monitoring, with check changes available only when you explicitly enable writes.

Hosted connector

Use the hosted Streamable HTTP endpoint in clients that support OAuth 2.1 and Dynamic Client Registration:

https://mcp.watchgoose.com/mcp

Available in

Watchgoose is available in the Claude connector directory.

Watchgoose is listed in ChatGPT plugins for GPT and Codex.

Watchgoose is listed in the official MCP Registry.

The watchgoose-mcp package is available on npm.

In Claude, open Settings > Connectors and add the endpoint as a custom connector. For Claude Code, run:

claude mcp add --transport http --scope user watchgoose https://mcp.watchgoose.com/mcp

The browser consent flow limits each connection to one project and defaults to read-only access. See the Watchgoose MCP documentation for the tool and privacy boundaries.

Requirements

• Node.js 20 or later
• A project-scoped Watchgoose API key from Project settings → API Access → Create (read-only)

Use an hcr_ read-only key when you only need check state and status history. Use an hcw_ read-write key when you also need pings, integrations, or check changes.

Claude Desktop

Add this entry to claude_desktop_config.json and restart Claude Desktop:

{
  "mcpServers": {
    "watchgoose": {
      "command": "npx",
      "args": ["-y", "watchgoose-mcp"],
      "env": {
        "WATCHGOOSE_API_KEY": "hcr_your_project_key"
      }
    }
  }
}

Claude Code

claude mcp add --env WATCHGOOSE_API_KEY=hcr_your_project_key \
  --transport stdio --scope user watchgoose -- npx -y watchgoose-mcp

Run claude mcp get watchgoose to check the connection.

Cursor

Create .cursor/mcp.json in your project, or add the same entry to your user MCP configuration:

{
  "mcpServers": {
    "watchgoose": {
      "command": "npx",
      "args": ["-y", "watchgoose-mcp"],
      "env": {
        "WATCHGOOSE_API_KEY": "hcr_your_project_key"
      }
    }
  }
}

VS Code

Create .vscode/mcp.json:

{
  "servers": {
    "watchgoose": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "watchgoose-mcp"],
      "env": {
        "WATCHGOOSE_API_KEY": "hcr_your_project_key"
      }
    }
  }
}

Run MCP: List Servers and start watchgoose.

Cline

Cline connects to the hosted endpoint with OAuth, so no API key is needed. Open MCP Servers > Configure MCP Servers and add this entry to cline_mcp_settings.json:

{
  "mcpServers": {
    "watchgoose": {
      "type": "streamableHttp",
      "url": "https://mcp.watchgoose.com/mcp"
    }
  }
}

Keep "type": "streamableHttp" exactly as written; without it Cline falls back to SSE and the connection fails. When Cline prompts you to authenticate, sign in to Watchgoose in the browser, choose one project and keep read-only access unless you want Cline to change checks. To use a project API key instead, add the npx entry from the Cursor section above to the same file.

Environment

The server classifies key prefixes locally and never probes the API to infer access. Setting WATCHGOOSE_ENABLE_WRITES=true cannot give a read-only key additional access.

Tools

Checks are addressed by stable 40-character unique_key values. UUIDs, ping URLs, and integration UUIDs are used only inside the server and are not shown to the AI client.

Integration assignments use exact integration names. Names must be non-empty and unique within the project.

create_check is a create-or-update operation when unique is supplied: a matching existing check has its supplied settings replaced. timeout without schedule also switches a scheduled check to interval monitoring: the stored schedule string remains, but no longer governs cadence. Include the confirmed schedule to retain scheduled monitoring. Other omitted settings remain unchanged. channels: [] clears all alert integration assignments. The tool therefore advertises destructiveHint: true. A new check awaits success/failure for completion-based cadence; a first failure can move it down. An optional start ping can initiate runtime monitoring before completion.

List results are capped at 100 checks, 100 pings, 200 flips, and 100 integrations. Every serialized tool result is also capped at 24,000 characters. Results include metadata when entries are omitted.

Security

• Writes are disabled by default, even with an hcw_ key.
• The API key is sent only in the X-Api-Key header and is never logged or returned.
• API redirects are rejected so credentials cannot be forwarded to another host.
• Check UUIDs, ping and update URLs, integration UUIDs, and unknown future API fields are removed from output.
• Pings keep only type, date, n, scheme, method, and duration. Source IP addresses, user agents, run IDs, and body URLs are removed.
• Ping bodies are never requested.
• The server has no telemetry.

Protect client configuration files that contain API keys. Prefer a dedicated project and the least-privileged key that exposes the tools you need. Revoke a key from Watchgoose project settings when it is no longer in use.

Report vulnerabilities privately as described in SECURITY.md.

Workflow skills

The repository includes portable skills for auditing scheduled-job coverage, drafting single-job setup, triaging missed heartbeats, reviewing existing monitoring health, and proposing approved create-only batches from an audit, list or CSV. See the skills guide for local installation, examples, and safety boundaries. Skills do not grant additional access, automatically run jobs, or change the existing directory listings.

The installable monitoring plugin lives in plugins/watchgoose-monitoring/, with one canonical skill tree and remote-MCP adapters. Load that subfolder rather than the repository root; the old root plugin/skill paths are superseded without automatic migration of installed copies. The npm CLI and hosted endpoint remain unchanged.

Each release also publishes the plugin as the root of pixeware/watchgoose-plugin:

• Gemini CLI: gemini extensions install https://github.com/pixeware/watchgoose-plugin
• Grok Build: grok plugin install pixeware/watchgoose-plugin --trust
• Any agent that reads Agent Skills: npx skills add pixeware/watchgoose-plugin, then add the hosted endpoint in your client

Development

npm install
npm run format:check
npm run lint
npm run typecheck
npm test
npm run publish:dry-run

Production integration tests are opt-in and require dedicated test-project keys. They never run in CI.

Every user-visible distribution shares one release version: the npm package, server.json, the hosted server, the Claude and portable plugin manifests and the OpenAI plugin. Bump WATCHGOOSE_RELEASE_VERSION in packages/core/src/constants.ts and the matching manifests together; tests/skills/release-version.test.ts fails if any of them disagree. The OpenAI upload ZIP is built with npm run build:openai-plugin.

Support

• Support: support@watchgoose.com
• Documentation: watchgoose.com/docs/mcp
• Privacy policy: watchgoose.com/legal/privacy
• Terms of service: watchgoose.com/legal/terms

License

MIT

Various badges