Skip to main content
hc
Submit
mcpSource-backedReview first Safety ✓ Privacy ✓

Railway MCP Server for Claude

Manage Railway projects, services, deployments, and environments from Claude — create projects, deploy templates, pull variables, trigger redeployments, and more — with the official Railway MCP server built into the Railway CLI.

by Railway·added 2026-06-18·
Claude CodeCodexCursorClaude Desktop
HarnessClaude CodeCodexCursorClaude Desktop
Review first — review before installing

Open the source and read safety notes before installing.

Safety notes

  • The Railway CLI MCP server runs as a local stdio process with full access to your authenticated Railway account — it can create, delete, and modify projects and services.
  • Use Railway's environment scoping and project permissions to limit blast radius; avoid running the MCP server under an account with access to production environments unless intentional.
  • A remote option (`--remote`) routes requests through `mcp.railway.com`; prefer local stdio if you prefer to keep credentials on-device.

Privacy notes

  • Railway commands may expose environment variable names and values from your projects into the Claude context; treat these as secrets.
  • The Railway CLI authenticates via stored tokens — keep your Railway account credentials secure.

Prerequisites

  • Railway CLI installed: `npm i -g @railway/cli` (or `brew install railway` on macOS).
  • Logged in to Railway: `railway login`
  • An MCP client such as Claude Code or Claude Desktop.

Schema details

Install type
cli
Troubleshooting
No
Source repository stats
Scope
Source repo
Collection metadata
Estimated setup
10 minutes
Difficulty
beginner
Tool listing metadata
Full copyable content
{
  "mcpServers": {
    "railway": {
      "command": "railway",
      "args": ["mcp"]
    }
  }
}

About this resource

Overview

The Railway MCP Server is the official Model Context Protocol integration built directly into the Railway CLI. It lets Claude manage your Railway infrastructure — creating projects, deploying templates, inspecting services, pulling environment variables, and triggering redeployments — through natural language. It runs as a local stdio process using your existing railway login session and is licensed under MIT.

A remote option is also available (railway mcp install --remote) which routes requests through https://mcp.railway.com instead of your local CLI.

Key capabilities

  • Project management — list, create, and inspect Railway projects.
  • Service operations — view services, check deployment status, trigger redeployments.
  • Environment variables — pull and inspect variables from any project environment.
  • Template deployment — deploy Railway templates from natural language.
  • Log inspection — retrieve recent service logs for debugging.
  • Environment management — list and switch between Railway environments.

How it compares

Server Cloud deployments Env var access Log retrieval Transport
Railway MCP Yes (Railway PaaS) Yes Yes stdio (local) or Remote
Fly.io MCP Yes (Fly.io) Yes Yes stdio
DigitalOcean MCP Yes (DO cloud) Limited Limited Remote
Vercel MCP Yes (Vercel) Yes Yes Remote
Render MCP Yes (Render) Limited Yes Remote

Railway's MCP is unique in bundling the integration into the CLI binary rather than shipping a separate npm package, meaning it always stays in sync with the Railway API.

Installation

Install the Railway CLI

# npm (macOS, Linux, Windows)
npm i -g @railway/cli

# macOS via Homebrew
brew install railway

Log in and configure Claude Code

railway login
railway mcp install --agent claude-code

Claude Desktop

{
  "mcpServers": {
    "railway": {
      "command": "railway",
      "args": ["mcp"]
    }
  }
}

Remote mode (optional)

To connect through Railway's hosted MCP endpoint instead of the local CLI:

railway mcp install --agent claude-code --remote

Requirements

  • A Railway account.
  • Railway CLI installed via npm i -g @railway/cli or brew install railway.
  • An MCP client (Claude Code or Claude Desktop).

Security

  • Runs as a local stdio process — credentials stay on-device in stdio mode.
  • Use environment scoping and project-level permissions to limit what Claude can access.
  • Remote mode (--remote) sends requests through mcp.railway.com.

Source Verification Notes

Verified on 2026-06-18:

  • Railway's MCP documentation at docs.railway.com/cli/mcp confirms the built-in MCP integration, railway mcp install --agent claude-code setup command, and both local stdio and remote transport options.
  • Railway's CLI installation guide at docs.railway.com/guides/cli documents npm i -g @railway/cli and brew install railway as the recommended install methods.
  • GitHub repo railwayapp/cli (MIT) is the official open-source Railway CLI codebase containing the MCP integration.
#railway#deployment#cloud#infrastructure#paas#devops

Source citations

Add this badge to your README

Show that Railway MCP Server for Claude is listed on HeyClaude. Paste this Markdown into your README — it renders the badge and links back to this page.

Listed on HeyClaude
[![Listed on HeyClaude](https://heyclau.de/badge/mcp/railway-mcp-server.svg)](https://heyclau.de/entry/mcp/railway-mcp-server)

How it compares

Railway MCP Server for Claude side by side with 3 alternatives on trust, install, platform support, and disclosed safety notes — all from reviewed registry metadata.

FieldRailway MCP Server for Claude

Manage Railway projects, services, deployments, and environments from Claude — create projects, deploy templates, pull variables, trigger redeployments, and more — with the official Railway MCP server built into the Railway CLI.

Open dossier 		Fly.io MCP Server for Claude

Manage Fly.io applications, machines, volumes, secrets, certificates, and organizations from Claude — with the official Fly.io MCP server built into the flyctl CLI.

Open dossier 		DigitalOcean MCP Server for Claude

Connect Claude to DigitalOcean — manage Apps, Droplets, managed Databases, Kubernetes, Container Registry, networking, and Functions — with DigitalOcean's official Model Context Protocol server.

Open dossier 		Render MCP Server

The official Render MCP server lets LLMs manage Render resources: create and manage web services, static sites, cron jobs, Postgres and Key-Value instances, monitor deploys, query logs and metrics, and run read-only SQL against Render Postgres.

Open dossier
Trust
Install riskReview firstReview firstReview firstReview first
Notes Safety ✓ Privacy ✓ Safety ✓ Privacy ✓ Safety ✓ Privacy ✓ Safety ✓ Privacy ✓
Categorymcpmcpmcpmcp
Sourcesource-backedsource-backedsource-backedsource-backed
AuthorRailwayFly.ioDigitalOceanrender-oss
Added2026-06-182026-06-182026-06-172026-06-11
Platforms
Claude CodeCodexCursorClaude Desktop
Claude CodeClaude Desktop
Claude CodeClaude Desktop
Claude CodeClaude Desktop
Source repo————
Safety notes✓The Railway CLI MCP server runs as a local stdio process with full access to your authenticated Railway account — it can create, delete, and modify projects and services. Use Railway's environment scoping and project permissions to limit blast radius; avoid running the MCP server under an account with access to production environments unless intentional. A remote option (`--remote`) routes requests through `mcp.railway.com`; prefer local stdio if you prefer to keep credentials on-device.✓The Fly.io MCP server runs locally with full access to your authenticated Fly.io account — it can create, delete, and modify apps, machines, and secrets. Fly.io warns that running the server remotely can give others access to run commands on your behalf; keep it bound to localhost unless you intend remote access. Destructive operations (machine deletion, secret updates) are available — review Claude's proposed commands before executing in production environments.✓Tools can create, update, restart, and delete live infrastructure (Apps, Droplets, Databases) — scope the API token and select only the --services you need. Destructive actions (delete, rollback) act on production resources; confirm before running them through Claude.✓Write-capable: tools can create and modify real Render infrastructure — create_web_service, create_static_site, create_cron_job, create_postgres, create_key_value, and update_environment_variables provision or change live resources that may incur billing. update_environment_variables replaces the complete environment variable set for a service; an incomplete array can drop existing variables. Created services run build and start commands you supply; treat generated commands as code execution on Render's platform. The server reaches Render's API over the network; the hosted option (https://mcp.render.com/mcp) sends your requests through Render's hosted MCP endpoint. Review and confirm tool calls before approving them, since an LLM can issue provisioning or env-var changes on your behalf.
Privacy notes✓Railway commands may expose environment variable names and values from your projects into the Claude context; treat these as secrets. The Railway CLI authenticates via stored tokens — keep your Railway account credentials secure.✓App names, machine IDs, secret names (not values unless explicitly requested), and log content may be surfaced into Claude's context. Fly.io API tokens (`FLY_ACCESS_TOKEN`) grant full account access — store them in your environment, not in repositories.✓Resource metadata, logs, and metrics enter the MCP client context and the model's prompt. The DIGITALOCEAN_API_TOKEN is a secret — store it in the client config or environment, never in shared repositories.✓Authentication uses a RENDER_API_KEY scoped to your Render workspace(s); anyone with the key can manage those resources. Keep it in a server-scoped header or server-scoped env block, not a top-level/global env block shared with other MCP servers. query_render_postgres runs SQL against your Render Postgres and returns row data to the LLM — query results may include sensitive application data. Logs and metrics tools (list_logs, list_log_label_values, get_metrics) surface application log contents and performance data to the model. update_environment_variables and service details can expose configuration values; avoid sending secrets you don't want the model to see. When using the hosted server, requests transit Render's hosted MCP infrastructure rather than staying entirely local.
Prerequisites
  • Railway CLI installed: `npm i -g @railway/cli` (or `brew install railway` on macOS).
  • Logged in to Railway: `railway login`
  • An MCP client such as Claude Code or Claude Desktop.
  • flyctl installed — see fly.io/docs/flyctl/install/ (Homebrew: `brew install flyctl`)
  • Logged in to Fly.io: `fly auth login`
  • Or set `FLY_ACCESS_TOKEN` environment variable for headless/CI use.
  • An MCP client such as Claude Code or Claude Desktop.
  • A DigitalOcean account.
  • A DigitalOcean API token (DIGITALOCEAN_API_TOKEN) with the scopes for the services you enable.
  • Node.js (npx) to run @digitalocean/mcp, or use the hosted remote endpoint.
  • An MCP client such as Claude Code or Claude Desktop.
  • A Render account
  • A Render API key created from Account Settings (dashboard.render.com/u/settings)
  • An MCP-compatible client (e.g. Claude Desktop, Cursor)
  • For the local binary only: the unzipped release executable, or Go to build from source
Install
railway mcp install --agent claude-code
fly mcp server --claude
claude mcp add digitalocean -e DIGITALOCEAN_API_TOKEN=<your-token> -- npx -y @digitalocean/mcp --services apps,droplets,databases
# Recommended: use Render's hosted MCP server (no local install required).
# Optional local binary — download from GitHub Releases, then point your
# MCP client at the unzipped executable:
#   https://github.com/render-oss/render-mcp-server/releases
Config
{
  "mcpServers": {
    "railway": {
      "command": "railway",
      "args": ["mcp"]
    }
  }
}
{
  "mcpServers": {
    "fly": {
      "command": "fly",
      "args": ["mcp", "server"]
    }
  }
}
{
  "mcpServers": {
    "digitalocean": {
      "command": "npx",
      "args": ["-y", "@digitalocean/mcp", "--services", "apps,droplets,databases"],
      "env": {
        "DIGITALOCEAN_API_TOKEN": "<your-token>"
      }
    }
  }
}
// Hosted server (recommended) — HTTP clients such as Cursor / Windsurf:
{
  "mcpServers": {
    "render": {
      "url": "https://mcp.render.com/mcp",
      "headers": {
        "Authorization": "Bearer <YOUR_API_KEY>"
      }
    }
  }
}

// Hosted server for Claude Desktop via mcp-remote; env is scoped to render only:
{
  "mcpServers": {
    "render": {
      "command": "npx",
      "args": [
        "mcp-remote",
        "https://mcp.render.com/mcp",
        "--header",
        "Authorization: Bearer ${RENDER_API_KEY}"
      ],
      "env": {
        "RENDER_API_KEY": "<YOUR_API_KEY>"
      }
    }
  }
}

// Local binary alternative (stdio transport); env is scoped to render only:
{
  "mcpServers": {
    "render": {
      "command": "/path/to/render-mcp-server",
      "args": ["--transport", "stdio"],
      "env": {
        "RENDER_API_KEY": "<YOUR_API_KEY>"
      }
    }
  }
}
Citations
ClaimUnclaimedUnclaimedUnclaimedUnclaimed

Signals

Loading live community signals…

More like this, weekly

A short, calm digest of reviewed Claude resources. Unsubscribe any time.