Levelrail MCP
Levelrail MCP gives an agent access to a self-hosted PaaS: apps, logs, deploys and rollbacks on servers running Levelrail.
- 207 ★ on the repository
- Last commit
- Apache-2.0
Broad access
Why this level
- The `standard` and `full` modes with `--allow-write` allow deploys, rollbacks and environment edits on live servers.
- A `root` token opens data for the whole fleet.
Install
- Works in
Manual install
npx skills add https://github.com/glincker/levelrail --skill levelrailThe coding-agent option from the README install table. The MCP server starts separately with `levelrail-mcp`.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The `standard` and `full` modes with `--allow-write` allow deploys, rollbacks and environment edits on live servers.
- A `root` token opens data for the whole fleet.
This is third-party code. Review the repository files before installing.
What it does
`levelrail-mcp` connects an agent to the Levelrail HTTP API, a self-hosted PaaS with `git push` deploys, HTTPS, logs and rollbacks. It can read apps, logs, status, domains and backups; deploy and rollback are enabled separately. By default a session lists a small set of about 2,400 tokens, and the remaining tool groups (over 240 in total) load on demand. The `read-only`, `standard` and `full` modes limit what the agent sees at all.
Who it is for. Owners of their own VPS and small servers who deploy with Levelrail and want to debug failures by talking to an agent.
Good fit when
- An app broke after a deploy, and the agent reads logs and failure context through `failure_context`.
- A domain does not open, and the agent runs `domain_doctor` to get a fix suggestion.
- Before a release the agent checks backup health and upgrade readiness.
- The agent joins Claude Code, Cursor, VS Code or Zed through a ready block from `--print-config`.
Not a fit when
- The server runs on macOS or Windows, while the Levelrail server part installs on Linux only.
- A `root` token is meant to run production with no confirmation.
- A mature 1.0 platform is required, while Levelrail stays on 0.x versions.
Example request
The api app stopped responding. Find the cause from logs and the last deploy, change nothing and propose a rollback.MCP
- Transport
- stdio, http
- Authentication
- API key
Environment variables
`APP_API_URL`required- Address of your Levelrail HTTP API.
`LEVELRAIL_TOKEN`required, secret- Access token. Its abilities decide which tools the agent sees.
Limitations
Levelrail is still on 0.x: the README treats a single server as production ready and multi-node as beta. The server part installs on Linux only, with just the CLI on macOS and Windows. The MCP server alone is useless without a running Levelrail and a token. The tool set follows token abilities, while `standard` and `full` modes with `--allow-write` permit deploys, rollbacks and other changes. For agents it is best to start with `--readonly`: the server refuses to start when the token has write abilities. Log text arrives in untrusted blocks, but the agent still reads application data.
How to disable. Remove the `levelrail` block from the client config and revoke the token in Levelrail. The PaaS itself keeps running.
FAQ
How do I get the client configuration?
Run `levelrail-mcp --print-config claude`. `cursor`, `vscode` and `zed` work instead of `claude`, and the block holds a placeholder instead of the real token.
How do I keep the agent read-only?
Start the server with `--readonly` or `--mode read-only`. The `--readonly` flag checks the token abilities and refuses to start when any include write.
How many tokens does the tool list cost?
About 2,400 tokens in the default mode. The other groups load with the `enable_toolset` command.
Where does the server part run?
On Linux only, on a VPS or bare metal. macOS and Windows get the CLI only.
What is progressive tool disclosure?
A session starts with a small core and three meta tools. `list_toolsets` shows the groups and `enable_toolset` loads the one you need.
Discussion
No questions or reviews yet.
Related
MCP serversA secure MCP runtime through which an agent operates local machines, servers and containers
The Railway hosting CLI: deploy and manage projects, plus one command that installs Railway skills and MCP for your agents
Self-hosted MCP server for running commands on remote servers over SSH, with tokens and roles
An MCP server and CLI that lets an agent query and provision cloud, SaaS and API resources with plain SQL
Official Heroku MCP server: apps, dynos, add-ons, pipelines, logs and Postgres through the Heroku CLI
Your machine's terminal, windows and desktop in a browser, plus an MCP server for agents to read and type into other sessions
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| Levelrail MCP (this card) | no | API key | 207 | |
| Coolify MCP | no | API key | 622 | exposes tools for Coolify; Levelrail MCP targets its own PaaS and narrows the set by token abilities |
| Dokploy MCP | yes | API key | 430 | manages Dokploy; Levelrail MCP is built on read-only, standard and full modes |
README in short
Levelrail is described as a self-hosted alternative to Vercel and Coolify, shipped as one Go binary with Docker and no Kubernetes. The agent section points to MCP, the permission model and a skill.