railway-mcp-edge-city

name	railway-mcp-edge-city
description	Query Railway logs, variables, services, and status for the Edge City control-plane and per-tenant Hermes sidecars via the Railway MCP. Use when debugging agentvillage card/digest generation, OpenRouter credits, or sidecar health and you need Railway data for a project OTHER than the default-linked Index project. Covers the snake_case project_id/environment_id gotcha and the known Edge City IDs.

railway-mcp-edge-city

The Railway MCP is pinned to one default linked project+environment (the Index Network Index project). Every tool silently ignores camelCase args and falls back to that default, so cross-project calls quietly return the wrong project's data.

The gotcha (this wastes the most time)

• Tool params are snake_case: project_id, environment_id, service_id — NOT projectId. Passing projectId is silently ignored → you get Index project data back.
• You must pass both project_id AND environment_id for a non-default project. With only project_id, the tool tries to reuse the linked env ID and fails: Failed to resolve environment: Environment "<linked-id>" not found.
• That error message lists the project's real environments — read it to get the right environment_id, then retry. (This is the fastest way to discover an env id.)

Edge City control-plane IDs (agentvillage)

• workspace Edge City = f4b23113-c71b-481f-83a5-c2ea3dde3b9a
• project agentvillage-controlplane = a57fc1a2-34fc-4e97-8056-4ea771ffb789
  • environment production = e4e4b158-701f-4f5a-8231-13e42ef8361f
  • service control-plane = e35802de-5319-4bc2-8f1f-3f838f934bc9
  • service dashboard-proxy = 269877b6-d8a8-4139-ae28-993e53479d4c, Postgres = 8b7aef42-a78e-4f4b-a262-7e018dbc53a8
  • per-tenant sidecars = services named hermes-<8hex> (non-hermes-pool-*); pool slots are unassigned.
• Verify/refresh IDs: railway_list_services { project_id } (snake_case!).

Recipes

List services / get logs / read env (always pass project_id + environment_id):

railway_list_services { project_id: "a57fc1a2-..." }
railway_get_logs      { project_id: "a57fc1a2-...", environment_id: "e4e4b158-...",
                        service_id: "e35802de-...", log_type: "deploy", lines: 200, search: "<tenant-id>" }
railway_list_variables{ project_id: "a57fc1a2-...", environment_id: "e4e4b158-...", service_id: "e35802de-..." }

Debugging agentvillage card/digest generation

The dashboard "Generate" button only queues the Edge — digest prepare cron on the tenant's Hermes sidecar (returns 202; never waits for the card). The card is produced by an OpenRouter-backed LLM session in that sidecar. So "pressed generate, no card" usually means the downstream sidecar run failed — not the button.

1. Get control-plane creds from its Railway vars: CONTROL_PLANE_API_KEY, public domain RAILWAY_PUBLIC_DOMAIN (→ CP_URL). Then follow the runbook packages/edge-city/agentvillage-controlplane/docs/OPENROUTER_WORKSPACE_TRIAGE.md:
   • GET $CP_URL/tenants → find tenant by email; confirm status:"live", sidecarReady:true.
   • GET $CP_URL/tenants/<id>?detail=1 | jq '.tenant.openrouter' → check disabled / limitRemainingUsd. A topup raises the scoped key limit but does NOT rotate the key or restart the container; if disabled:true persists after topup, escalate (don't keep adding spend).
2. For raw failure reasons, pull the tenant sidecar logs: find its hermes-<hex> service in railway_list_services, then railway_get_logs it. Control-plane logs only show cron list/trigger warnings (cron trigger failed for tenant <id>), not the LLM/staging-script error.

Note: these env vars contain live secrets — never paste keys into public channels.

See also: audit-digests (digest card quality/ownership audit).