@cronwatch/mcp

An MCP server for apps that use @cronwatch/sdk. It talks to the JSON API the app mounts with cw.routes(), so Claude Code, Cursor or any MCP client can list jobs, read a failing run's error and output, run a check, and silence a job during a fix. Every port serves the same API at the same paths, so it works against those apps too: the Ruby gem's Cronwatch::Web, the Python package's cw.routes() (or cronwatch.django), the PHP package's $cw->routes() (and its Laravel, Symfony, WordPress, Drupal and Craft CMS integrations), the Go module's cw.Routes(), the Rust crate's cw.routes(options), nested in axum or served as a tower service, the Elixir package's Cronwatch.Web, forwarded to from a Phoenix router, the Java library's cw.routes(), served by its Spring Boot starter, a servlet filter or the JDK's own server, and the .NET package's cw.Routes(), mapped with app.MapCronwatch() in ASP.NET Core. Its end-to-end tests drive it against each port's dashboard as well as the SDK's, when CRONWATCH_TEST_RUBY, CRONWATCH_TEST_PYTHON, CRONWATCH_TEST_PHP, CRONWATCH_TEST_GO, CRONWATCH_TEST_RUST, CRONWATCH_TEST_ELIXIR, CRONWATCH_TEST_JAVA or CRONWATCH_TEST_DOTNET is set.

claude mcp add cronwatch \
  -e CRONWATCH_URL=https://app.example.com/cronwatch \
  -e CRONWATCH_TOKEN=your-token \
  -- npx -y @cronwatch/mcp

Any stdio MCP client:

{
  "mcpServers": {
    "cronwatch": {
      "command": "npx",
      "args": ["-y", "@cronwatch/mcp"],
      "env": { "CRONWATCH_URL": "https://app.example.com/cronwatch", "CRONWATCH_TOKEN": "your-token" }
    }
  }
}

Tools

| Tool | | |---|---| | list_jobs | every job with health, schedule, last run and next due | | get_job | one job with recent runs: errors, output tails, metrics | | run_check | look for missed and stuck runs now and send due alerts | | silence_job, unsilence_job | pause and resume alerts for a job | | forget_job | remove a job that no longer exists in the code | | get_setup_guide | the code to add CronWatch to a job |

Give the token through CRONWATCH_TOKEN, as above. The --url and --token flags still work, but a flag is visible to anyone who can list processes (ps), so keep the token in the environment. Node 22 or newer.

list_jobs, get_job and get_setup_guide are marked read-only; silence_job and forget_job are marked destructive, so clients that ask before destructive tools will ask. A run's error and output are written by the job, so tool results label them as untrusted data for the model to read, not follow.

The token grants everything the dashboard can do. Docs: cronwatch.dev/docs/mcp. MIT.