Claude Railway connectorWhat Claude can do in your Railway account.
The Claude Railway connector exposes 16 tools: 8 read, 7 write, 1 undocumented. Published by Railway, it lets Claude check your projects and logs, deploy a GitHub repository, redeploy a service, handle feature flags and hand complex jobs to Railway's own agent.
Verified Trustpilot reviews · AI, automation & growth agency
What the Railway connector changes for a developer
Railway hosts apps and their infrastructure; following a deploy usually means moving between the dashboard, the service and its logs. With this connector, you ask Claude in the conversation: it reads the environment status, the deployments and the logs itself, and it can also create a project, deploy a GitHub repository, redeploy a service or change a feature flag, within the projects your Railway account can reach.
Debug a failed deploy. get-status shows the environment, its services and the status of their latest deployments, along with any staged work, and get-logs pulls the build, deploy and HTTP logs Claude reads through to find the cause.
Ship a new service from GitHub. From a GitHub repository connected to Railway, create-deployment creates the service and starts its first deploy; list-deployments then tracks it.
Change a feature flag after reading it. get-feature-flag reads a flag before set-feature-flag changes its default, as Railway itself advises; Railway labels that second tool destructive.
What it will not do: nothing runs on its own when a deploy breaks; Claude acts only while it answers you. Redeploying cannot give a service its very first deployment either. We have no n8n, Make or Zapier page for Railway; our Integrations hub lists the platforms we cover, and the Claude connectors page gathers the others.
Five words before you connect
The Claude vocabulary used on this page, in one minute.
- Connector
- The link you set up once between Claude and an account you already own, so Claude can work on it while it answers you.
- Tool
- One named thing a connector lets Claude do on that account. Claude picks the tools it needs, and the directory sheet lists them by name.
- Authorization
- Railway's own sign-in screen, where you hand Claude the access it will use afterwards. Granted once per person, taken back the same way.
- Approval
- The confirmation Claude waits for before it goes ahead with something that changes the account, shown in the chat at the right moment.
- MCP
- The shared standard connectors are built on: it is what lets an assistant like Claude talk to an outside service such as Railway.
Connect the Claude Railway connector in three steps
- 01
Find Railway in Connectors
Open Claude's settings, go to Customize then Connectors, and search for Railway. On a Team or Enterprise workspace, an Owner or Primary Owner has to enable it for the organization before members can sign in to it individually.
- 02
Connect and sign in to Railway
Click Connect on the Railway row, then log in through the window Railway opens. If the link breaks later, Disconnect and connect again; you can also withdraw the access from your Railway account itself.
- 03
Approve Railway's screen
Read Railway's authorization screen before you validate it: Railway, not Claude, decides what the access covers. Afterwards, Claude only reaches the projects and workspaces your own Railway account can already open.
The 16 tools of the Railway connector
Railway gives Claude 16 tools: 8 that read your account, 7 that change something in it, and 1 no official source describes.
Eight tools read, seven change your infrastructure, and one has no description anywhere. Names stay exactly as Claude shows them.
- 8 read
- 7 write
- 1 not documented
What Claude reads (8)
8 toolsEight tools that inspect your account, projects, deployments, logs and flags without touching them.
get-feature-flag
Fetches a single feature flag, which Railway calls a Signal, by name, either from a project or from its parent workspace.
get-logs
Reads Railway logs across several streams: runtime, build, HTTP proxy requests, network flows and DNS lookups, for one deployment or for a service's recent deployments.
get-status
Reports the state of a project environment: its services and their latest deployment status, replicas, cron schedules, volumes, buckets, and any staged work not yet live.
list-deployments
Lists a project's recent deployments, newest first, with who triggered each one, its regions, and whether it can be redeployed or rolled back. Filters by environment, service or status are available.
list-feature-flags
Lists the feature flags of a project, optionally with those of its parent workspace. Project flags are editable with admin access; workspace flags stay read-only from a project.
list-projects
Lists every Railway project your account can access, which gives Claude the names and IDs it needs for any other request.
list-services
Returns the IDs and names of all services and environments inside one Railway project, a lightweight map before a deeper look.
whoami
Returns the profile of the Railway user whose account the connector uses.
What Claude changes (7)
7 toolsSeven tools that create, deploy or reconfigure. Railway labels five of them destructive.
accept-deploy
Asks you firstCommits every staged change in a Railway environment and triggers a deploy of the result. Railway flags it as destructive and tells Claude to use it only once you have explicitly confirmed you want to deploy.
create-deployment
Approval: see the ruleCreates a new service from a GitHub repository you have connected to Railway and starts its first deployment. Railway instructs Claude not to guess a repository: it asks which one, in owner/name form.
create-project
Approval: see the ruleOpens a new, empty Railway project in your account, ready to receive services.
delete-feature-flag
Approval: see the ruleDeletes a feature flag scoped to a project. Flags defined at workspace level cannot be removed from a project. Railway marks the tool as destructive.
railway-agent
Approval: see the ruleSends a message to Railway's own AI agent for complex, multi-step infrastructure work; that agent can inspect services, diagnose failures and take actions on your behalf. Railway marks it destructive.
redeploy
Approval: see the ruleRe-runs the latest deployment of a service in an environment, reusing its existing build, or an earlier deployment's build if Claude names it. It only works on a service that has already deployed there.
set-feature-flag
Approval: see the ruleCreates a project-level feature flag or sets its default value, and can replace its targeting rules. Railway advises reading existing flags first.
Undocumented (1)
1 toolThe directory lists this name, but no official text says what it does, and we will not guess from the name.
deploy-artifact
This name appears in the directory's tool list and nowhere else: Railway's published list of tools leaves it out, and no official page explains it. So this page says nothing about what it does.
What Claude asks before it acts
By default, Claude stops and asks before each action it carries out on an account for someone. Railway's directory sheet adds that destructive actions are gated behind explicit confirmation, without saying which tools that covers. Only accept-deploy carries, in its own description, the instruction to wait for your explicit go-ahead.
On Team and Enterprise workspaces, owners decide whether a member may let some actions through without being asked again. They can also restrict what a connector may do for the whole organization, for example keep reading and block writing, and no member overrides that from their own account. Claude works with the rights of the person who connected it and no more: Railway scopes everything to the projects and workspaces that person can already reach.
Which Claude plans include it
Anthropic's general rule: remote connectors are open to all users on Claude, Cowork, Claude Desktop and mobile, while desktop extensions install in Claude Desktop.
On Team and Enterprise, an Owner or Primary Owner enables a connector for the organization before members can sign in to it. For Railway's current status, check its sheet in the official directory, linked from the button at the top of this page. No official source publishes plan availability connector by connector, so that sheet is the only place to confirm it.
What the Railway connector does not do
Only Railway documents this connector, through its directory sheet and the tool list its server publishes: neither the Claude help center nor Anthropic's developer docs have a page about it.
A connector is not an automation: Claude calls its tools during a reply, and nothing fires on its own when a deploy fails. The directory's verification badge is not a security audit, and Anthropic says on every sheet that it does not control the tools a publisher exposes. The sheet also describes publishing small artifacts to a public URL through Railway Functions, without naming the tool that does it.
Need help connecting Railway to Claude?
A person reads every message.