Production service returning 403 "host_not_allowed" from outside my home network

Question

Hi team,

My production service is blocking requests from almost every IP except my own home network. This is stopping Anthropic's Claude from connecting to it via MCP, and other external integrations have stopped working too. I'm not technically experienced — I had an AI assistant help me investigate, and we narrowed it down to something on Railway's side, not in my code.

**My service**

* Project: `b-brain-hub-production`
* Public URL: `https://b-brain-hub-production.up.railway.app`
* No custom domain, no TCP proxy — just the default `.up.railway.app` domain on port 8080.
* Stack: Node 20 + Express. No IP-allowlist code or middleware in the app (we grepped the codebase and dependencies).

**What's happening**  
From most networks, every URL on my service returns:

```
HTTP/2 403
```

`x-deny-reason: host_not_allowed`

`content-type: text/plain`

`content-length: 21`

`Host not in allowlist`

The 403 response is missing your usual edge headers (`server: railway-edge`, `x-railway-request-id`, `x-railway-edge`) — but successful responses include all of them. The header name `x-deny-reason` and the body `Host not in allowlist` do not appear anywhere in my code, dependencies, or environment variables.

**Confirmed IP dependence (same minute, same paths)**

* From my home Wi-Fi: `GET /api/health-check` → `HTTP/2 200` with full Railway headers; request id `XV6nN-ErTBm8LC-7jq4OvQ`.
* From a sandbox VM on a different ASN: every path returns the 403 above.
* From my phone on cellular (Wi-Fi off): also unreachable.

Affected paths include:

* `POST /mcp/v2`
* `GET /.well-known/oauth-protected-resource`
* `GET /.well-known/oauth-authorization-server`
* `GET /api/health-check`

**Reproduction**

```
curl -i -X POST -H "Content-Type: application/json" -d '{}' \
```

` https://b-brain-hub-production.up.railway.app/mcp/v2`

**Reference IDs for log correlation**

* Successful request id (from my home IP): `XV6nN-ErTBm8LC-7jq4OvQ`
* Earlier successful request id: `fFvjkGU3Sx60LY2v6WHkDg`
* \- Error references from Claude when its connector tried to reach the service: `ofid_9b2bae343c845fa2` and `ofid_3201da116f50a22b` (most recent attempt, just now)
* My current outbound IP (the one that's getting through) is on the Fastly/Apple-relay range — happy to share the exact address privately if it helps your team correlate logs.

**My questions**

1. What is emitting the `x-deny-reason: host_not_allowed` header? Is it a Railway edge feature, a beta access-control or anti-abuse rule, or a proxy in front of the edge?
2. Is there a setting in my project that turned this on? If so, where in the dashboard can I manage it?
3. The MCP endpoint needs to be reachable from Anthropic's servers (Claude's backend), not just from my browser. Can you check whether their IP ranges are being blocked, and tell me how to allow them?

Thank you!