Railway MCP Server

# Fix list: Railway MCP Server

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/railway-mcp-server, the October 2026 research run, assessed 9 October 2026. Grade C, 56.7 out of 100.

This is everything the published grade says the listing lacks, the biggest possible gain to the total first. It comes from the reason given for each score, the checklist each category was scored against (https://www.anchorterminal.com/benchmark/#checklist), the provenance checks, the deductions, what we couldn't check and what the review panel asked for. A fix counts at the next check, once it's public.

For a coding agent working on Railway MCP Server: work through the items below in the product, its docs and its public pages. Each category gives the reason for its score, with the points each checklist item earned, and the checklist itself, so the gap is the items that earned less than their points. Change the product, not the wording, and keep a note of what you changed and where it's published.

## 1. Reliability, 60 out of 100, up to 8 more on the total

Why it scored 60: Scored as a hosted server. status.railway.com is an in-house status page with component history by region and for the API and OAuth login (20). From 11 July to 8 October 2026 it lists about 25 incidents, on a page that says it reports only incidents with wide impact. Most are slow builds and deployments. They include services and databases unavailable for some users in US West on 3 October, API degradation that paused Hobby deployments on 29 September, a five-minute routing failure for all domains on 30 September and a hardware fault in US West on 29 August. Durations are not shown in the history, and the MCP server is not a component (10 of 30). No limit is stated for the MCP server. The public API's limits are published with numbers, 100 to 10,000 requests an hour by plan (8 of 15). The API docs give `Retry-After` on 429, backoff advice and retry rules for mutations, with no idempotency key (10 of 15). A 99.99% SLA is stated for Enterprise, with terms by contract (5 of 10). The docs carry no beta label and the registry entry is 1.0.0, but the April 2026 launch note called it ready for public testing and no later statement was found (7 of 10).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-reliability):

Hosted APIs, MCP servers, models and platforms.

- 20, a public status page with component history (Statuspage, Instatus, BetterStack or the vendor's own).
- 0 to 30, the incident record for the last 90 days on that page. 30 for a clean record or trivial incidents only, 20 for minor incidents only, 10 for one major outage (an hour or more of a core API down, or errors across the board), 0 for several. 5 when there's no history we could read, and the note says so.
- 15, rate limits documented with numbers.
- 15, documented 429 or overload handling (Retry-After, backoff guidance), and idempotency keys or safe-retry guidance where writes are involved.
- 10, an SLA published for any paid tier.
- 10, the surface agents use is generally available, not beta or preview.

Local packages, SDKs, frameworks and stdio MCP servers.

- 20, installs from an official package with supported runtimes stated.
- 25, a public CI and test suite, passing on the default branch.
- 0 to 25, open crash or regression issues relative to activity (25 for few and handled, 0 for many, old and unanswered).
- 15, semver discipline and breaking changes called out in a changelog.
- 15, version 1.0 or later, or declared stable.

Protocols are read from their reference implementations, the public facilitators or servers, spec stability and test vectors.

## 2. Payments & pricing, 40 out of 100, up to 7.5 more on the total

Why it scored 40: No x402, MPP or L402 on the server (0). Railway can be provisioned through Stripe's Agentic Provisioning Protocol, which is a separate route and earns nothing here. Plan fees and per-second prices for CPU, memory, volumes and egress are public (20). The trial gives $5 of credit for 30 days with no card, followed by a Free plan with $1 a month (20). Dynamic client registration needs no setup, but a person has to sign in and approve the grant in a browser or with a device code (0).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-payments):

The published rubric, also on the [x402 page](https://www.anchorterminal.com/x402/).

- 40, a machine payment protocol (x402, MPP or L402) on the tool's own endpoints. 10 to 30 when it covers only some endpoints or only goes through a third party, and the note says which.
- 20, per-call or per-unit pricing published without a login. 10 for public plan-only pricing, 0 for "contact sales" or prices behind a login.
- 20, a free tier or trial that doesn't need a card.
- 20, autonomous onboarding, meaning an agent can get access without a person signing up in a browser (keyless use, x402, a programmatic key API).

Payment platforms and agent wallets rarely charge for their own API over a machine protocol, so the first line has steps for them, and the highest one that applies counts. 40 when x402, MPP or L402 runs on all their own endpoints, 30 when it runs on part of their own API, 25 when their merchants can accept one, 20 for running a facilitator, 15 for paying as a buyer, and 0 when the only protocol is their own. Merchant acceptance sits above a facilitator because the platform's own customers can charge agents through it, while a facilitator settles for sellers who wire up the protocol themselves. The counter-argument (a facilitator does more for the protocol as a whole) has a point. Each note says which step applied.

Open-source software you run yourself is scored on its hosted or paid option if it has one. A free, self-hosted package with nothing to buy gets 20, 20 and 20 for the last three lines, and 0 to 40 for the first only if it ships a payment protocol.

## 3. Schema & documentation, 56 out of 100, up to 7.2 more on the total

Why it scored 56: The hosted server's source is closed and `tools/list` needs a signed-in client, so no tool schema was read. A server card and a registry entry are public, and the CLI proxy source shows that tools declare their parameters (15 of 25). llms.txt on both sites and a Markdown twin of every docs page (10). The docs name the 11 tools and describe four of them in a line each. The descriptions the server sends were not read (8 of 20). Input types, enums and bounds could not be read (5 of 15). The docs give example prompts and the 401 challenge, and the API docs list error codes. No list of tool errors was found (9 of 15). Railway's changelog is weekly and dated, but the growth from seven tools to 11 has no dated entry found, and the hosted server shows a fixed version 1.0.0 (9 of 15).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-schema):

APIs and MCP servers.

- 25, a machine-readable contract (a public OpenAPI file or similar; for MCP, typed JSON Schema inputs on every tool).
- 10, llms.txt or Markdown docs served for agents.
- 0 to 20, descriptions that say what a tool is for, when to use it and when not to, read from the tool definitions in the source or the API reference.
- 0 to 15, typed inputs with enums, constraints and required fields, and no free-form JSON blobs.
- 0 to 15, examples and documented error responses.
- 15, versioning and a public changelog.

Models are read from the API reference, the OpenAPI file, llms.txt, the structured-output and tool-use docs and the model cards. Frameworks from docs a model can follow, typed interfaces, examples and the API reference.

## 4. Agent ergonomics, 59 out of 100, up to 6.7 more on the total

Why it scored 59: Eleven tools, one over the ten that earns full marks, with `railway-agent` taking multi-step work in a single call (20 of 25). The three list tools are the only read tools apart from the flag tools. Pagination and size controls could not be read (6 of 20). A missing or expired token answers 401 with a `WWW-Authenticate` challenge, the CLI proxy returns a login hint, and API errors carry a code and a trace id (12 of 20). The docs say destructive tools are marked at the protocol level so clients prompt. Read-only hints were not established, and there is no idempotency key (12 of 20). The CLI proxy fills project context from the linked directory for tools that accept it. Two official SDKs for the platform API were not found (9 of 15).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-ergonomics):

- 0 to 25, context cost. For MCP, the number and size of the tool definitions (25 for ten or fewer compact tools, 15 for 11 to 30, 5 for more than 30, plus up to 10 back for toolsets, dynamic loading or read-only subsets). For APIs, whether responses can be sized (field selection, limits, summaries).
- 20, pagination, filtering and output-size controls.
- 20, actionable, documented error responses, codes and messages an agent can recover from.
- 20, idempotency or safe retries, and for MCP the `readOnlyHint` and `destructiveHint` annotations.
- 15, sensible defaults, few required parameters, and official SDKs in at least two languages.

Models are read for tool use, structured output, prompt caching, context length, batch and SDKs. Frameworks for how much code and how many defaults a tool-calling agent with MCP needs.

## 5. Security & auth, 65 out of 100, up to 6.1 more on the total

Why it scored 65: OAuth with PKCE required, dynamic client registration, grants limited to the workspaces and projects a person picks, one-hour access tokens and rotating refresh tokens. Revocation is in the dashboard only, and the MCP resource lists a single scope, `workspace:member` (25 of 30). There is no read-only grant for the hosted server. Destructive tools carry hints that prompt in clients that respect them, and project tokens are refused (10 of 20). `railway-agent` reads logs and configuration, which are untrusted content. The docs tell users to review actions and keep to non-critical environments, with no prompt-injection guidance found (4 of 15). Workspace audit logs record the actor and can be exported through the API, kept 48 hours on Free and Hobby, 30 days on Pro and 18 months on Enterprise. No marker for MCP calls was found (11 of 15). SOC 2 Type II and SOC 3, a HIPAA BAA, a Bugcrowd bounty that is invite-only and public incident reports. No security.txt (15 of 20).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-security):

- 0 to 30, the credential model. 30 for OAuth 2.1 with scopes, or scoped and revocable keys with rotation. 20 for plain revocable API keys. 10 for one all-powerful key. 10 off when a secret can travel in a URL query string as a documented option.
- 0 to 20, read-only or least-privilege modes, and confirmation or approval for destructive actions.
- 0 to 15, prompt-injection posture where the tool returns untrusted content (documented mitigations or guidance). A tool that returns no untrusted content gets 10.
- 0 to 15, audit logs or per-call visibility for the operator.
- 0 to 20, a security programme. security.txt or a disclosure policy, a bug bounty, SOC 2 or ISO 27001, advisories handled in public.

Models are read for retention, whether API data trains models (and whether that's off by default), zero-retention options and certifications. Frameworks for telemetry defaults, approval hooks, guardrails and sandboxing.

## 6. Transparency & trust, 73 out of 100, up to 2.4 more on the total

Made of editorial 61, provenance 84.

Why it scored 73: The hosted server is closed source under clear terms. The CLI proxy and the local server are MIT (20 of 30). The DPA is published in full. The privacy policy's full text was not read. Audit log and image retention are stated by plan, and nothing was found on what the hosted server or `railway-agent` keeps from a session (18 of 30). The old npm server was deprecated on 23 May 2026 with a shim and a registry notice, and API fields are marked `@deprecated` before removal with no stated period. The terms let Railway suspend or discontinue any part of the Services (10 of 20). The DPA names Google Cloud Platform and the United States, promises ten days' notice of a new sub-processor and points to a list at trust.railway.com that was not read (13 of 20).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-transparency):

- 0 to 30, source availability and licence clarity. 30 for open source under an OSI licence, 15 for closed with clear terms, 0 for unclear terms.
- 0 to 30, data handling and retention statements that agree with each other (privacy policy, DPA, retention periods, subprocessors).
- 0 to 20, a deprecation policy or notices with dates.
- 0 to 20, telemetry disclosed with an opt-out (local software), or subprocessors and data locations disclosed (hosted).

The other half of Transparency and trust is the provenance score, computed from checked facts (below). The category score is the mean of the two.

Provenance checks not met in full (half of this category, computed from checked facts):

- Terms of service: published, but our reader couldn't read it (7 of 10)
- Privacy policy: published, but our reader couldn't read it (7 of 10)
- security.txt: not found (0 of 10)

## 7. Maintenance & community, 83 out of 100, up to 1.5 more on the total

Why it scored 83: The CLI that carries the proxy reached v5.64.1 on 8 October 2026 and Railway's changelog has an entry dated 1 October (30). More than 100 CLI tags in the last 90 days and a weekly changelog (20). GitHub's API refused us for its rate limit, so issue handling was not read. Railway runs a feedback thread for the server on Central Station (10 of 25). The official MCP registry lists `com.railway/mcp` as active under the vendor's domain namespace (15). The CLI has CI, a cargo-audit workflow and a lockfile. The workflow results were not seen (8 of 10).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-maintenance):

- 0 to 30, time since the last release, or the last published model or API change for a closed service. 30 within 30 days, 20 within 90, 10 within 180, 0 older.
- 20, at least three releases or dated changelog entries in the last 90 days.
- 0 to 25, responsiveness. Issues and pull requests answered on GitHub (the open issues and how recent the replies are). For closed services, a public changelog and a support or community channel that answers, 0 to 15.
- 15, presence in the official MCP registry under a verified namespace (MCP servers), or current official SDKs (APIs and models).
- 10, package health, current dependencies and CI.

Models are read for deprecation notice periods and model churn rather than release counts.

## Deductions

Each comes off the total. A fixed and documented problem counts for less at the next check.

- 30 March 2026. A CDN configuration change cached GET responses on about 0.05% of Railway domains for 52 minutes, so pages meant for one user may have been served to another. Railway reverted it, purged the cache and published a report the same day. Fixed and documented, so 4 is deducted (https://blog.railway.com/p/incident-report-march-30-2026-authenticated-user-data-cached).

## What we couldn't check

What we couldn't read counted as absent. Publishing it on a page a plain HTTP fetch can read (not only in a browser) lets the next check count it.

- unchecked: the hosted server's tool schemas, descriptions and annotations. The source is closed and `tools/list` needs a signed-in client, so Schema and Agent ergonomics are scored from the docs.
- unchecked: the full text of the Privacy Policy. Its Markdown twin is a summary, and the terms forbid scraping pages of the Services, so the HTML page was not read.
- unchecked: the sub-processor list at trust.railway.com, for the same reason.
- unchecked: GitHub stars and issue handling for railwayapp/cli. GitHub's API refused us for its rate limit.
- unchecked: whether the public API's rate limits apply to MCP tool calls, and whether audit logs mark MCP calls.
- unchecked: the terms of the Enterprise 99.99% SLA, which are not published.
- unchecked: incident durations. The status history lists titles and closing notes without start and end times.
- The Terms of Service were read once from the HTML page before the clause against crawling or scraping was seen. After that only Railway's published agent files, Markdown twins, well-known files and the status pages were requested. robots.txt on railway.com, docs.railway.com, status.railway.com and blog.railway.com signals ai-input=yes. This matters before our probes run.
- The lead's description (create projects, deploy templates, manage environments, pull variables) comes from the docs' opening line. The hosted tool list has `create-project` and `redeploy` but no template, environment or variable tool.
- The launch note of 17 April 2026 called the server ready for public testing. No later statement that it is generally available was found, and the docs carry no beta label.
- The registry name is `com.railway/mcp` while Railway's server card says `com.railway/railway`.
- `lastRelease` is the date of the CLI tag v5.64.1, which carries the proxy. The hosted server has no dated releases.

## Weaknesses

- The hosted server's source and tool schemas are not public, so inputs can only be read after signing in.
- The MCP resource lists one scope, `workspace:member`, with no read-only grant.
- The docs say the server can deploy templates, manage environments and pull variables, but no hosted tool is named for those. They go through `railway-agent` or the local server.
- status.railway.com lists about 25 incidents from 11 July to 8 October 2026, most on builds and deployments.
- Railway reported a 52-minute CDN caching error on 30 March 2026 that may have shown one user's pages to another.

## What costs an agent a turn today

The notes we give agents before they call it. Each one is a workaround an agent shouldn't need.

- Connect to `https://mcp.railway.com` with OAuth, or run `railway mcp` after `railway login`. Project tokens are refused.
- Treat `redeploy`, `accept-deploy`, `delete-feature-flag` and `railway-agent` as actions that change production. Confirm the project and environment first.
- `railway-agent` spends model tokens billed to the workspace. Use the single-purpose tools for listing and redeploying.
- For variables, domains, volumes, logs or metrics as separate tools, use `railway mcp local`, which has a different tool set.
- On a 401, follow the `WWW-Authenticate` challenge and run the OAuth flow again. Access tokens expire after one hour.

## When it's done

Send what changed and where it's published as a dispute (https://www.anchorterminal.com/builders/#disputes, or `POST https://www.anchorterminal.com/api/v1/contact` with `"kind": "dispute"`). Disputes are answered in public, and the listing is checked again by the same checklist. Paying for an audit or a listing claim changes nothing here.