Connect the AI assistant on your laptop, such as Claude Code, Claude Desktop, Cursor or Codex, to your rdev panel. It can then answer questions like "is any of my servers in trouble?", "why did the last task fail?" or "which agent is waiting for me?", and act on a deploy or an agent session when you ask it to.
This is the panel's own MCP server at https://rdev.cloud/mcp. It is separate from the MCP access your agents have on the server itself (see AI agents): that one manages sites, databases and services inside the environment, this one shows your account across all your servers.
What the assistant can do
| Tool | What it does |
|---|---|
list_servers |
Your servers with online status, rdev version, projects, CPU, memory and disk use, and alerts (offline, restart needed, security updates, disk almost full, failed services) |
server_details |
One server's services, PHP and Node.js versions, processes, scheduled tasks with their last exit code, and backup storage location |
agent_sessions |
The latest agent sessions and their state: working, waiting for you, finished or stopped on an error |
recent_tasks |
The latest tasks the panel sent to your servers and how they ended; it can list only the failed ones |
deploy_requests |
Team deploys waiting for approval, in the teams you own or administer |
approve_deploy, reject_deploy |
Decide on a waiting deploy, with the same rules as the Team page: owners and administrators only, within an hour of the request |
stop_agent_session |
Stop a running agent session, as the Agent sessions page does |
The assistant sees only the servers you can change in the panel: your own, and on a Team plan the main environment if you own or administer the team. It never gets license keys, env values, encryption passwords or tokens. The assistant asks you before it approves, rejects or stops anything, as for any MCP tool that changes something.
Create a token
- Open Settings > Connected apps in the panel.
- Under Connect an AI assistant, give the token a name you will recognise later, for example "Claude Code on my MacBook", and choose Create token.
- Copy the command or the configuration shown under it. The token is shown only once; if you lose it, create a new one.
Each token is listed under Connected apps with the time it was last used. Disconnect revokes it at once. Use one token per assistant and laptop, so you can revoke one without touching the others.
Treat the token like a password: anyone who has it can see your servers and approve deploys. Keep it out of project files that are committed to a repository.
Claude Code
Run the command from the panel in a terminal on your laptop:
claude mcp add --transport http --scope user rdev https://rdev.cloud/mcp --header "Authorization: Bearer <token>"
--scope user makes the server available in every project. Check the connection with claude mcp list or /mcp inside Claude Code.
Cursor and other MCP clients
Most clients read the same JSON. In Cursor, add it to ~/.cursor/mcp.json (all projects) or .cursor/mcp.json in a project:
{
"mcpServers": {
"rdev": {
"type": "http",
"url": "https://rdev.cloud/mcp",
"headers": {
"Authorization": "Bearer <token>"
}
}
}
}
The server speaks MCP's Streamable HTTP transport and authenticates with the Authorization: Bearer header, so any client that supports remote servers with custom headers works.
Codex
Put the token in an environment variable, for example export RDEV_MCP_TOKEN=<token> in your shell profile, and add this to ~/.codex/config.toml:
[mcp_servers.rdev]
url = "https://rdev.cloud/mcp"
bearer_token_env_var = "RDEV_MCP_TOKEN"
Claude Desktop
Claude Desktop's custom connectors sign in with OAuth, which the panel does not offer yet. Connect it through the mcp-remote bridge instead (it needs Node.js on your laptop). In Settings > Developer > Edit Config, add:
{
"mcpServers": {
"rdev": {
"command": "npx",
"args": ["-y", "mcp-remote", "https://rdev.cloud/mcp", "--header", "Authorization: Bearer ${RDEV_MCP_TOKEN}"],
"env": { "RDEV_MCP_TOKEN": "<token>" }
}
}
}
Then restart Claude Desktop.
Limits
An assistant can make up to 120 requests a minute; one question usually takes a few. A revoked or mistyped token gets 401 Unauthorized; create a new token and update the client's configuration.