Type at least two letters to search every page of the docs.

↑ ↓ to navigate ↵ to open esc to close

Using rdev

MCP tools

Every agent on your server comes connected to rdev through MCP (Model Context Protocol), the standard way for an AI agent to use tools. So the agent does not only write code: it sets up what the code needs, checks that it runs and fixes what breaks, without asking you to open a terminal or the panel.

Why it matters

  • The agent finishes the job. After writing a queued job it adds the queue worker; when a feature needs Redis or a database it installs and connects them; when composer asks for a PHP extension it installs it. You get a working feature, not a list of steps to do by hand.
  • It debugs with facts, not guesses. The agent reads the end of the Laravel log, the last error, a worker's output and the mail your app sent, then fixes the cause.
  • Risky changes are reversible. The agent takes a checkpoint before a large change and can restore it.
  • Every agent gets the same tools. Claude Code, Codex, opencode, Qwen Code, GitHub Copilot CLI, Gemini CLI, Cursor CLI and Antigravity CLI all get the same MCP tools and the same rules, so you can switch agents without setting anything up again.
  • Nothing leaves your server. The MCP server runs inside your environment and talks to the agent directly. Your code, logs and data are never sent to rdev.cloud.
  • You stay in control. Claude Code asks you before it uses a tool for the first time, as for any MCP tool. If your license runs out, tools that add projects or sites or install new versions are refused; everything else keeps working.

What the agent can do

Area Tools What they do
Projects rdev_project_add, rdev_project_service Add a project (empty, from a Git repository or from a recipe) and give it services such as MySQL, PostgreSQL, Redis, Meilisearch or S3 storage, with the .env filled in
Sites create_site, link_site, list_sites, set_site_secure, add_domain, add_proxy Create and link .dev.test sites, turn on HTTPS, add domains and proxies to dev servers
PHP list_php, install_php, set_site_php, set_default_php, install_php_extension, set_php_setting Install PHP versions, pick one per site, add extensions such as mongodb, change php.ini settings
Other languages rdev_node_*, rdev_lang_*, rdev_system_install Install Node.js, Python, Ruby, Go, .NET, Java or Flutter versions, pick one per project, install system packages a library needs
Databases and services create_database, list_databases, list_services Create databases and see which services run
Processes and schedules rdev_jobs, rdev_process, rdev_cron, rdev_job_logs Add, change, restart and remove queue workers, Horizon, Reverb and scheduled tasks, and read their output
Logs and mail rdev_app_log, list_mails, get_mail, list_dumps Read the end of the app log or its last error, the mail the app sent, and dump() output
Errors rdev_errors Read a project's unresolved errors from local error tracking
Checkpoints rdev_checkpoint List, take and restore checkpoints of a project
Previews rdev_preview Share a site as a temporary, password-protected preview when you ask
Health rdev_backup_status, rdev_license_status, diagnose, status Check the nightly backup, the license and the environment

rdev_app_log returns the end of a project's application log, or with last_error only the last error and its stack trace. It works for Laravel out of the box (the newest file in storage/logs). It also reads Symfony's var/log/dev.log, WordPress's wp-content/debug.log (turn on WP_DEBUG and WP_DEBUG_LOG in wp-config.php; the tool tells you how if the file is missing) and Rails' log/development.log. For projects whose dev server rdev runs (Node.js, Python, Ruby, Go, .NET, Java and others), it reads the dev server's output, where last_error recognizes Python tracebacks, JavaScript, Java, .NET and Ruby stack traces and Go panics. The same is available on the command line: rdev project logs <name> --app or --last-error. In a browser on your tailnet, https://rdev-logs.dev.test shows the same for every project and refreshes as new lines arrive.

Processes and scheduled tasks and checkpoints are explained on their own pages.

Examples

Ask in plain words; the agent picks the tools:

  • "Add a newsletter signup that sends a welcome email through the queue." The agent writes the code, adds a queue worker, sends a test signup and reads the captured email.
  • "The checkout page returns 500, fix it." The agent reads the last error in the log, fixes it and checks the page again.
  • "Start a new Laravel project called shop with PostgreSQL and Redis." The agent adds the project, the services and the site.
  • "Show this to the client." The agent starts a preview link and gives you the address and password.

Setup

None. The installer and rdev agents install register rdev's MCP server (rdev mcp) with every agent they set up. An agent you install yourself can use it too: add rdev mcp as a local (stdio) MCP server in its settings and point it to AGENTS.md.

The panel's MCP server

There is a second, separate MCP server at https://rdev.cloud/mcp for the AI assistant on your laptop. It shows your account across all your servers (health, agent sessions, failed tasks, deploys waiting for approval) instead of working inside one environment. See Connecting an AI assistant.