Provides tools for interacting with Railway's GraphQL API, allowing AI agents to manage Railway projects and deployments using paired account tokens.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@railguey-mcp-cloudflare show me my Railway projects and recent deployments"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Railguey in the cloud. A Cloudflare Worker MCP server that bridges Grok to Railway GraphQL.
The CLI lives in eidos-agi/railguey. This repo is the hosted Streamable HTTP MCP that Grok can actually add as a connector — Railway’s own MCP (mcp.railway.com) is OAuth/CLI-shaped and cannot be added to Grok directly.
Live worker: railguey.eidos-agi.workers.dev
MCP: https://railguey.eidos-agi.workers.dev/mcp
Product page: eidosagi.com/tools/railguey
Repo | What it is |
Project-scoped Railway CLI (Go). Local shells, GitHub Actions, agents with a workspace path. | |
| Hosted MCP server (this Worker). Grok talks HTTP here; the Worker talks Railway GraphQL. |
The Worker script is still named railguey on Cloudflare. That name is the workers.dev host and must match wrangler.toml name. The GitHub repo is named for what it is.
Related MCP server: @crisog/railway-mcp-server
Railway tokens never go through Grok chat. Pairing is a device-code flow:
Grok calls account_pair_begin (or a code is minted here).
You open /pair/<CODE> on this worker.
Paste a Railway token and give the account a slug (eidos, personal, client-acme).
The worker validates the token against Railway, stores it in KV, and Grok sees it via account_list.
Codes expire in 10 minutes, are single-use, and burn after 8 failed pastes.
Account tokens (railway.com/account/tokens), workspace tokens, and project tokens (Project → Settings → Tokens) are all accepted. Kind is recorded so Grok knows the scope.
Each pair is a named slot. Railway tools take an optional account slug. One slot is the default.
This matches the CLI model (~/.railguey/accounts.json) — named accounts, one default — stored in Workers KV instead of a home directory.
GET / — status page
GET /health — JSON health (no secrets; includes paired account metadata)
GET /pair — enter a claim code
GET|POST /pair/:code — paste a Railway token
POST /mcp — MCP Streamable HTTP (auth required)
MCP_AUTH_TOKEN — bearer Grok sends as Authorization: Bearer …
KV ACCOUNTS (railguey-accounts) — paired Railway tokens, claim codes, default slug
Never commit tokens. wrangler.toml only has the KV namespace id and Cloudflare account id.
grok.com/connectors → Add connector → Other
Server URL: https://railguey.eidos-agi.workers.dev/mcp
Header: Authorization: Bearer <MCP_AUTH_TOKEN>
Ask Grok to pair a Railway account, then open the claim URL it returns
git push to main is the deploy. This repo is permanently attached to the live Worker:
Push to eidos-agi/railguey-mcp-cloudflare main (especially index.js).
Grok automation railguey-git-deploy fires on that push, uploads index.js to Worker railguey, and keeps KV + MCP_AUTH_TOKEN.
Hourly backup redeploys main if a push event is missed.
No Cloudflare API token in GitHub. No Cloudflare GitHub App. Changing MCP behavior means editing index.js and pushing.
loader.js is a Dynamic Worker Loader (git-pull at the edge). It is not live — this Cloudflare account is on the free Workers plan (API 10195). wrangler.toml main stays index.js.
name = "railguey" must not change. Renaming the Worker would break the live host and Grok’s connector URL.
MIT
This server cannot be deployed
Host your MCP tool over streamable HTTP in one command.
Remote MCP server for supportsheep: run AI interviews and manage support content for your blog.
MCP server for AI agents to plan, verify, and deploy Cloudflare-native apps.
Talk to your public-facing AI from any MCP client — Claude, ChatGPT, Cursor, Cline, Windsurf.
