--- title: "Railway plugin for Grok, manage CDN caching in the CLI" date: 2026-07-16T21:39:00.000Z number: 0299 url: https://railway.com/changelog/2026-07-17-railway-plugin-grok --- # Railway plugin for Grok, manage CDN caching in the CLI Your agents multiply, and Railway follows them around. The official Railway plugin lands in the Grok Build plugin marketplace this week, so xAI's coding agent can deploy and operate your projects out of the box. The CLI also picks up CDN cache controls with the new `railway cdn` command. Let's get into it! 🚄 # Railway plugin for Grok Build [Installing the Railway plugin from the Grok Build marketplace and deploying with it](https://cms.railway.com/media/Grok-Plugin-Hello-World.mp4) The official Railway plugin is now listed in the [Grok Build plugin marketplace](https://github.com/xai-org/plugin-marketplace). If you use xAI's coding agent, run `grok`, and if your prompt includes Railway you'll see a hint that allows you to one click install the plugin. You can also run `/marketplace`, select xAI official, then **railway**, and press `i` to install. The plugin packages everything the agent needs to work on your Railway projects: - The `use-railway` skill, which routes deploys, service configuration, environments, networking, and troubleshooting through the Railway CLI. - Railway's hosted MCP server at `mcp.railway.com` - A hook that auto-approves Railway CLI and API calls, so the agent isn't interrupted for permission on every command. Grok Build joins Claude Code, Codex, and Cursor on the list of agents with a first-class Railway plugin, all shipped from the [railway-skills](https://github.com/railwayapp/railway-skills) repo. Before the marketplace listing, installing the plugin in Grok meant pointing it at the repo manually. It's browsable and one keypress away now. Try it out, and let us know what you think on [Central Station](https://station.railway.com/new?type=feedback). # Manage CDN caching in the CLI ![CDN cache management from the Railway CLI](https://cms.railway.com/media/image-2.png) In case you missed it: Railway has a [built-in CDN](https://docs.railway.com/networking/cdn). Enable it on a service and responses are cached at the edge, close to your users, with no third-party setup. We wrote about how we built it in [How to build a 30M RPS CDN in 30 days with Rust and WASM](https://blog.railway.com/p/railway-cdn). Before this, cache settings and purges lived in the dashboard. Now the CLI exposes the same controls via the new `railway cdn` command: - `railway cdn status` reports whether caching is enabled for the selected service, along with the HTML caching mode, default TTL, stale-while-revalidate state, purge-on-deploy policy, and the last purge times. - `railway cdn enable` and `railway cdn disable` toggle caching, matching the dashboard behavior. Enabling requires an applied public domain. - `railway cdn update` changes settings on a service that already has caching enabled: `--html-caching `, `--default-ttl`, `--swr`/`--no-swr`, and `--purge-on-deploy `. - `railway cdn purge ` clears the HTML cache or the full cache on demand. Every command supports the standard `--service`, `--environment`, and `--project` selectors, plus `--json` so you can purge the cache from a deploy script or let an agent tune TTLs for you. Update with `railway upgrade`, read the [CDN docs](https://docs.railway.com/networking/cdn), run `railway cdn --help`, and let us know what you think on [Central Station](https://station.railway.com/new?type=feedback). # Fixes and improvements - We shipped project token support for feature flags. `railway flag` resolves its target project from your CLI context (a linked project or `RAILWAY_TOKEN`), and project tokens can create and update flags. Previously the command resolved the project through the linked workspace, and managing flags required a personal login, which kept CI jobs and agents out. * We fixed `railway config pull` flattening single-region replica maps into a plain region list. Pulled configs previously lost your placement intent, so pushing the same file back could reshape where replicas run. Region maps now survive the round-trip unchanged.