OPERATIONS / STATUS

Check the runtime directly.

Check Working Machines runtime health, MCP availability, expected authorization responses, and safe troubleshooting steps.

Production runtime

The canonical hosted runtime is https://app.workingmachines.dev. Its public health endpoint is https://app.workingmachines.dev/health. Use that endpoint for a current, machine-readable availability check instead of relying on a cached status claim on this page.

A healthy runtime confirms that the Working Machines service can answer requests. It does not guarantee that every external provider is available, that a specific account connection is still authorized, or that a requested action is permitted by workspace policy.

  • Runtime: https://app.workingmachines.dev
  • Health check: https://app.workingmachines.dev/health
  • Developer documentation: https://www.workingmachines.dev/docs

MCP availability

The MCP endpoint is protected. A request without valid credentials should receive an authorization challenge, so a missing-authorization response is expected and does not by itself mean the service is unavailable.

MCP clients should begin with the public discovery document, follow the advertised OAuth protected-resource metadata, and complete authorization before listing or running tools. Do not place access tokens in URLs, support messages, or prompts.

  • MCP endpoint: https://app.workingmachines.dev/mcp
  • Discovery: https://www.workingmachines.dev/.well-known/mcp
  • Agent instructions: https://www.workingmachines.dev/agents.md

How to isolate a failure

First check the runtime health endpoint. If it responds, determine whether the failure happened during discovery, authorization, connection selection, policy validation, or provider execution. The structured error response should identify the failing layer and whether a retry is safe.

For provider-specific failures, confirm that the selected connection belongs to the intended workspace, the provider grant contains the required scopes, and the upstream application is operational. Read actions may be safe to repeat; do not automatically repeat a write action unless its result and idempotency behavior are known.

  • 401 or authorization challenge: complete the advertised OAuth flow
  • 403 or policy rejection: review identity, scope, and workspace policy
  • Provider error: inspect the upstream status and granted scopes
  • Timeout: verify the execution record before retrying a mutation

Report a problem safely

When requesting help, include the time of the failure, the public endpoint involved, the action name, and a sanitized request or execution identifier. Never include OAuth tokens, API keys, provider secrets, personal data, or complete tool results in a public report.

Use the contact page for support paths and the security page for responsible disclosure guidance. Public documentation, discovery resources, and the OpenAPI contract remain available even when an authenticated workspace request cannot be completed.