Live Charlotte Area Transit System (CATS) bus and light rail data via three read-only MCP tools.
Find a specific vehicle or route's vehicles (find_vehicle): look up by vehicle number (e.g. 2301, LRV307) or route (e.g. 9, 501, Blue Line, Mt. Holly Road), with optional bus/train filter. Returns GPS position, heading, speed, occupancy, headsign, and next scheduled stop.
List all in-service vehicles (list_vehicles): get current GPS coordinates for every bus and train, filterable by mode or route, with a limit (default/cap 250) plus countsByMode and totalInService.
Get arrival predictions at a stop (get_arrivals): query by stop id (02400), stop code, or partial stop name (CTC Station), optionally filtered by route or mode. Returns minutes away, predicted and scheduled times, schedule deviation, vehicle number and its live position, plus service alerts when available.
Access live CATS GTFS-Realtime feeds (vehicle positions, trip updates, alerts refreshed every 20s) joined to static schedule data, with features like exact-first route matching, no negative ETAs, stale-data fallback with feedAgeSeconds, and ISO 8601 UTC times/WGS84 coordinates.
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., "@cats-mcp When is the next bus arriving at CTC Station?"
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.
QueensCoach is an MCP server for live Charlotte Area Transit System (CATS) bus and light rail data, built on the agency's public GTFS-Realtime feeds. It runs over stdio, launched by the MCP client that uses it, or over HTTP with Google OAuth in front of it, for a hosted server. Both transports serve the same three tools.
A public instance runs on Google Cloud Run. Sign in with any Google account:
https://queenscoach.adamwanninger.com/mcpThere is zero guarantee of uptime. The hosted server is provided as-is. It may be
slow, down, switched off by its spending cap, or retired without notice. For anything
you rely on, run your own: over stdio, or on your own Google Cloud project
with deploy/GCP.md.
Signing in tells the server your Google account's email address, which is used only to decide whether to admit you, and is never stored. The tokens it issues record your account's opaque Google ID and nothing else about you. The privacy policy and terms of service cover the hosted server.
Claude calls a remote MCP server a connector. Custom connectors are available on Claude's paid plans.
Open Settings → Connectors. On the web that's claude.ai/settings/connectors; in the desktop app, Settings then Connectors.
Click Add custom connector at the bottom of the list.
Give it a name, QueensCoach, and paste the URL above as the remote MCP server
URL. Leave the advanced OAuth fields empty: this server registers your client
automatically.
Click Add, then Connect on the connector that appears. A browser window opens for the Google sign-in; approve it and it closes itself.
In a chat, open the tools menu and check that QueensCoach is enabled. Its three tools then appear.
The connector belongs to your Claude account, so it follows you across web, desktop, and mobile. To disconnect, remove it from that same Connectors page; that revokes the tokens this server issued.
claude mcp add --transport http queenscoach https://queenscoach.adamwanninger.com/mcpThen run /mcp, pick queenscoach, and choose Authenticate, which opens the same
Google sign-in. /mcp shows the connection's state afterwards. A server added this way
loads when Claude Code next starts.
Any MCP client that supports remote servers over streamable HTTP with OAuth works: give it the same URL and it discovers the rest.
Related MCP server: OC Transpo MCP Server
Tool | Purpose |
| Locate one bus/train by vehicle number, or every vehicle on a route, and return GPS coordinates. |
| Current GPS coordinates of every bus and train in service. |
| Estimated arrival times at a specific stop or station. |
find_vehicleArgument | Type | Notes |
| string | Vehicle number as shown on the bus/train, e.g. |
| string | Route to locate: |
|
| Optional filter. |
At least one of vehicle or route is required. Returns position, heading, speed,
occupancy, headsign, and the next scheduled stop.
list_vehiclesArgument | Type | Notes |
|
| Optional filter. |
| string | Optional single-route filter. |
| integer | Max vehicles to return (default and cap: 250). |
Includes countsByMode and totalInService so the total is visible even when the
list is truncated.
get_arrivalsArgument | Type | Notes |
| string | Required. Stop id ( |
| string | Optional route filter. |
|
| Optional filter. |
| integer | Max arrivals (default 10, cap 50). |
Returns minutes away, predicted and scheduled times, schedule deviation, the vehicle
number, and that vehicle's live position. When a name query is ambiguous, the best
match is used and the runners-up are listed under otherStopsMatchingQuery. Service
alerts affecting the stop or its routes are attached when present.
Requires Python 3.11+.
python3 -m venv .venv
.venv/bin/pip install .Pick one with --transport or QUEENSCOACH_TRANSPORT; the default is stdio.
queenscoach # stdio (default)
queenscoach --transport http --port 8000 # streamable HTTP + Google OAuthFor a server the client launches itself. No authentication: the client already owns the process.
Register it with Claude Code:
claude mcp add queenscoach -- /absolute/path/to/queenscoach/.venv/bin/queenscoachOr in an MCP client config file:
{
"mcpServers": {
"queenscoach": {
"command": "/absolute/path/to/queenscoach/.venv/bin/queenscoach"
}
}
}python -m queenscoach runs the same server, so any interpreter with the package
installed works as the command.
stdout carries MCP protocol traffic only; all diagnostics go to stderr.
For a hosted server anyone with the URL can reach. Every request to /mcp needs a
bearer token, and the only way to get one is to sign in with a Google account that is
on the allow list.
How the sign-in works. MCP clients register themselves dynamically and expect an authorization server at the MCP server's own origin. Google offers neither dynamic registration nor tokens audience-restricted to a third-party resource, so this server is its own OAuth 2.1 authorization server and delegates only the login to Google:
MCP client <--OAuth--> queenscoach <--OAuth--> GoogleGoogle's answer is used exactly once, to learn which account signed in. That email is checked against the allow list, and only then does this server mint its own tokens. Google's tokens are never handed to the client.
One-time setup in Google Cloud. At console.cloud.google.com/auth/clients, create an OAuth client of type Web application and add one authorized redirect URI:
https://your-public-url/auth/google/callbackIt must match QUEENSCOACH_PUBLIC_URL exactly. The server logs the URI it expects at startup.
Copy the client ID and secret into the environment below.
Run it. .env.example lists every setting; the shell form is:
export QUEENSCOACH_GOOGLE_CLIENT_ID=...apps.googleusercontent.com
export QUEENSCOACH_GOOGLE_CLIENT_SECRET=...
export QUEENSCOACH_ALLOWED_EMAILS=you@example.com
export QUEENSCOACH_PUBLIC_URL=https://queenscoach.example.com
queenscoach --transport http --port 8000Then point a client at https://queenscoach.example.com/mcp; it discovers the rest and opens
a browser for the Google sign-in. In Claude Code:
claude mcp add --transport http queenscoach https://queenscoach.example.com/mcpAccess is denied by default. Startup fails unless QUEENSCOACH_ALLOWED_EMAILS,
QUEENSCOACH_ALLOWED_DOMAINS, or an explicit QUEENSCOACH_ALLOW_ANY_GOOGLE_ACCOUNT=true says who
may get in, so a misconfigured deployment is unreachable rather than open to every
Google account on the internet. Unverified Google addresses are always refused.
Endpoints.
Path | Purpose |
| The MCP endpoint. Requires |
| Points clients at the authorization server. |
| This server's OAuth metadata. |
| Dynamic client registration (RFC 7591). |
| The OAuth endpoints. |
| Where Google returns the user. |
scripts/install.sh does a whole deployment: a system
user under /opt, a Cloudflare tunnel and its DNS record created over the API,
both systemd units, and a verification pass. No port forwarding, so it works
behind CGNAT or a locked router. See deploy/.
scripts/deploy-gcp.sh does the same on Google Cloud
Run, in your own GCP project: the project itself, Firestore for sign-ins, the
client secret in Secret Manager, a container built by Cloud Build, a monthly
budget with an optional hard spend cap, and the same verification pass. It scales
to zero, so a personal server costs next to nothing. See
deploy/GCP.md.
Deployment notes.
By default the server speaks plain HTTP and expects a tunnel or proxy to
terminate TLS, which is what the install script sets up. Setting
QUEENSCOACH_TLS_CERT and QUEENSCOACH_TLS_KEY instead makes it serve HTTPS itself, for a
deployment with nothing in front of it.
QUEENSCOACH_PUBLIC_URL is what clients dial and is this server's OAuth issuer
identifier, so it must be the external URL, not the bind address.
Token state is in memory by default and therefore per-process: restarting
invalidates outstanding tokens. QUEENSCOACH_TOKEN_STORE=firestore keeps it in
Firestore instead (install the gcp extra: pip install 'queenscoach[gcp]'), so
sign-ins survive restarts and every instance shares them. Pair it with
QUEENSCOACH_STATELESS_HTTP=true so that any instance can answer any request.
Access tokens last an hour and refresh tokens 30 days, both rotated on refresh.
Realtime (GTFS-Realtime protobuf, refreshed every 20s):
https://gtfsrealtime.ridetransit.org/GTFSRealTime/Vehicle/VehiclePositions.pb
https://gtfsrealtime.ridetransit.org/GTFSRealTime/TripUpdate/TripUpdates.pb
https://gtfsrealtime.ridetransit.org/GTFSRealTime/Alert/Alerts.pb
Static schedule (cached 6h), used to turn feed identifiers into route names, stop names, and coordinates:
https://gtfsrealtime.ridetransit.org/GTFSStatic/api/GTFSDownload/GTFS.zip
Only routes.txt, stops.txt, and trips.txt are read; stop_times.txt and
shapes.txt are the bulk of the archive and are not needed.
Verified against live feed captures:
VehiclePosition.stop_id and current_stop_sequence are unusable. None of the
158 vehicle stop ids in a sample capture matched any stop in the published schedule,
and reported sequence numbers exceeded the trip's own stop count (e.g. sequence 192
on a 52-stop trip). This server never surfaces them; next-stop data comes from the
TripUpdates feed instead, whose stop ids resolve 100%.
StopTimeEvent.delay is never populated. Schedule deviation is computed from
time minus scheduled_time, which are both present.
TripUpdates cover ~83% of active vehicles, so nextStop is omitted rather than
guessed for the remainder.
Route matching is exact-first, so a query of 5 returns route 5, not 501 or 510.
Arrival predictions already in the past are filtered out; no negative ETAs.
Feed responses are capped in size and time-bounded; one slow feed cannot hang a call.
Concurrent calls share a single in-flight fetch per feed, and one call giving up does not abort a fetch the others are awaiting.
If a refresh fails but cached data exists, the last good data is served rather than
an error. feedAgeSeconds on every response shows how stale it is.
The alerts feed is supplementary: if it fails, get_arrivals still returns arrivals.
Times are ISO 8601 UTC; coordinates are WGS84 decimal degrees.
All optional; defaults target the CATS feeds above. Durations are in milliseconds.
Variable | Default |
| CATS vehicle positions feed |
| CATS trip updates feed |
| CATS alerts feed |
| CATS static GTFS zip |
|
|
|
|
|
|
|
|
|
|
Feed URLs must be http or https; anything else is rejected at startup.
Variable | CLI | Default |
|
|
|
Read only when --transport http is selected.
Variable | CLI | Default | Notes |
|
|
| Bind address. |
|
|
| Bind port. |
|
|
| External origin; the OAuth issuer. |
| required | From Google Cloud credentials. | |
| required | From Google Cloud credentials. | |
| — | Allowed addresses, comma- or space-separated. | |
| — | Allowed bare domains, e.g. | |
|
| Opt in to admitting every Google account. | |
|
| — | PEM chain, to serve HTTPS directly. |
|
| — | PEM private key. Required with the above. |
|
| Access token lifetime. | |
|
| Refresh token lifetime. | |
|
|
| |
|
| Firestore database for the token store. | |
|
| Serve without MCP sessions, for restarts and multiple instances. |
One of the three allow-list settings is required; see above.
Module | Role |
| Environment parsing and validation |
| Bounded, time-limited HTTP fetch |
| TTL cache with single-flight refresh |
| GTFS-flavored CSV reading |
| Static schedule: routes, stops, trips |
| GTFS-Realtime protobuf decoding |
| Domain layer: joins realtime to schedule, resolves queries |
| The three tools' behavior and JSON payloads |
| MCP tool registration and schemas |
| OAuth authorization server, with Google as the login |
| Where OAuth state is kept, and the in-memory default |
| The Firestore token store (the |
| Streamable HTTP transport and the Google callback route |
| CLI entry point and transport selection |
Plus scripts/install.sh, which deploys the HTTP
transport onto a Debian host, and scripts/deploy-gcp.sh
with the Dockerfile, which deploy it to Google Cloud Run.
.venv/bin/pip install -e '.[dev]'
.venv/bin/pytest # offline, against recorded feed fixtures
.venv/bin/mypy # strict
.venv/bin/ruff check .
.venv/bin/ruff format .Tests run against protobuf and GTFS fixtures captured from the live feeds, so they are
deterministic and make no network calls. tests/test_feed_http.py is the exception: it
serves canned responses from a loopback socket so the byte cap and timeout are exercised
for real.
tests/test_http.py drives the whole OAuth handshake against the real ASGI app -
registration, /authorize, the Google callback, /token, then an authenticated
tools/list - with Google's token endpoint replaced by a stub, so no account or network
is needed.
tests/test_token_store.py runs every token-store test against both stores. The
Firestore half needs the emulator (gcloud emulators firestore start, then set
FIRESTORE_EMULATOR_HOST) and is skipped without it. tests/test_stdio.py starts the
real stdio server in a child process; CI also runs it against a plain pip install .,
to prove stdio needs none of the optional extras.
MIT - see LICENSE.
Locate a specific CATS bus or train and return its current GPS coordinates. Give "vehicle" for a vehicle number (e.g. "2301"), or "route" to get every vehicle currently running a route (e.g. "9", "501", "Blue Line"). Includes heading, speed, occupancy, and next scheduled stop when available.
| Name | Required | Description | Default |
|---|---|---|---|
| mode | No | ||
| route | No | ||
| vehicle | No |
| Name | Required | Description |
|---|---|---|
No output parameters | ||
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already declare readOnlyHint=true and openWorldHint=true. The description adds useful behavioral context by specifying the return data (GPS coordinates, heading, speed, occupancy, next scheduled stop) and noting that the stop is included 'when available'. It does not contradict annotations, and the mention of 'when available' aligns with the open-world hint. However, it does not disclose behavior such as what happens when no match is found or whether both vehicle and route are allowed.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is two sentences, front-loaded with the primary purpose and immediately followed by usage examples. Every word contributes to understanding the tool's behavior and parameters. There is no redundant or filler content.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
The description covers the main usage patterns (vehicle- and route-based lookup) and mentions the output fields. However, it does not specify behavior when both vehicle and route are provided, when neither is provided, or how the mode parameter applies. Given the existence of an output schema, the missing parameter-precedence details are a gap that could lead to incorrect usage.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
The input schema already provides descriptions for all three parameters (mode, route, vehicle). The description adds extra meaning by explaining the mutually exclusive use of vehicle vs. route ('Give vehicle for a vehicle number... or route to get every vehicle'). It does not clarify the mode parameter or interaction between fields, so the added value over the schema is limited.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description starts with a specific verb ('Locate'), names the resource ('a specific CATS bus or train'), and states the output ('current GPS coordinates'). It also distinguishes two search modes (by vehicle number or by route) and clarifies the scope ('every vehicle currently running a route'). This clearly differentiates it from siblings like list_vehicles and get_arrivals, which likely list all or handle scheduled times.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description gives parameter usage instructions ('Give vehicle for a vehicle number... or route to get every vehicle') but provides no guidance on when to choose this tool over list_vehicles or get_arrivals. It does not mention exclusions, alternatives, or when the tool is not appropriate, leaving the agent to infer tool selection from the name alone.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Estimated arrival times of buses or trains at a specific stop or station. Accepts a stop id, stop code, or part of a stop name. Reports minutes away, schedule deviation, the vehicle number, and any service alerts for that stop.
| Name | Required | Description | Default |
|---|---|---|---|
| mode | No | Only show bus or train arrivals. | |
| stop | Yes | Stop id, stop code, or part of a stop name, e.g. "02400" or "CTC Station". | |
| limit | No | Maximum arrivals to return. | |
| route | No | Only show arrivals for this route. |
| Name | Required | Description |
|---|---|---|
No output parameters | ||
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
The annotations already declare readOnlyHint and openWorldHint; the description helps confirm these by saying 'Estimated'. It adds no operational limitations beyond what the schema provides, and it doesn't mention any rate limits, pagination, or fallback/ambiguous name handling. It does add 'service alerts' as a kind of output, which is lightly useful, so the description is adequate but not deep.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
Two sentences capture the main purpose, input flexibility, and key fields. The most important information is placed first. No filler is present.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
For a read-only tool with four parameters fully documented in the schema and an output schema present, the description provides enough context for an agent to know what the tool is for, what it accepts, and what it returns. It does not need to restate schemas because those are adjacent.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
The schema has 100% coverage of four parameters, each with its own description. The text like 'stop id, stop code, or part of a stop name' mainly restates the schema's stop description. It adds no extra meaning for limit, mode, or route, so meaning comes primarily from the schema itself.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description names the resource ('estimated arrival times of buses or trains') and the specific action ('at a specific stop or station'), making the tool's function explicit. It also lists the input formats and useful output fields, which differentiates it from sibling tools find_vehicle and list_vehicles.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
It implies when to use the tool (when you need arrival times at a stop), but it never states an explicit 'when-not-to-use' or contrasts with siblings. No exclusions are given, so an agent can infer intent, but someone selecting among the three siblings must rely on whatever context the user provides without extra guidance.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Return the current GPS coordinates of every CATS bus and train in service. Optionally filter to buses or trains, or to a single route.
| Name | Required | Description | Default |
|---|---|---|---|
| mode | No | ||
| limit | No | Maximum vehicles to return. | |
| route | No | Restrict results to one route. |
| Name | Required | Description |
|---|---|---|
No output parameters | ||
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already declare readOnlyHint=true and openWorldHint=true, so the safety profile is established. The description adds the behavioral fact that it returns 'current' coordinates and only vehicles 'in service,' which clarifies the data scope. It does not go beyond that (e.g., mention of latency, rate limits, or pagination), so it makes a modest addition beyond the annotations.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is two sentences with no waste. The core action is front-loaded ('Return the current GPS coordinates...'), and the optional filters are mentioned compactly. Everything written earns its place, making it highly efficient for an agent to parse.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
An output schema is present, so return format details are not needed in the description. The tool is a simple listing operation with optional filters; the description covers the essence, and the schema covers parameters. The only minor gap is the lack of explicit mention of potential limits (like pagination), but the limit parameter and output schema mitigate this. Overall, it is sufficiently complete for an agent to invoke correctly.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is 67%, meaning most parameters already have meaning in the schema (mode, limit, and route each have descriptions). The description repeats these filters ('buses or trains', 'single route') without adding extra semantics like value formats or edge cases. Since the schema already covers the parameters, the baseline of 3 is appropriate.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description states a specific verb ('Return'), a clear resource ('GPS coordinates of every CATS bus and train in service'), and explicitly distinguishes this from a single-vehicle lookup ('every'). It also mentions optional filters, which makes the tool's scope unmistakable and separates it from the sibling find_vehicle.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description says 'Optionally filter to buses or trains, or to a single route,' which gives clear context on how to narrow results. However, it does not explicitly state when to prefer this tool over siblings like find_vehicle (e.g., 'for all vehicles, use this; for a specific one, use find_vehicle'). The 'every' phrasing implies bulk listing but does not name alternatives or exclusions.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
v1.0.0find_vehicleget_arrivalslist_vehiclesScored across 3 tools
find_vehicle and list_vehicles overlap significantly: both can return every vehicle on a route, making the boundary between them unclear. get_arrivals is distinct, but the route-listing capability in find_vehicle duplicates list_vehicles with a route filter.
All tools use a consistent verb_noun snake_case pattern: find_vehicle, list_vehicles, get_arrivals. The verbs are distinct and clearly indicate the action, with no mixed conventions.
Three tools is a reasonable, focused set for a real-time transit information server. Each tool covers a distinct core need—vehicle location, fleet listing, and arrivals—without unnecessary bloat.
The tools cover the essential real-time transit operations: locating vehicles, listing vehicles, and fetching arrivals. Minor gaps exist, such as no dedicated route or stop directory tool, but the provided inputs (route numbers, stop ids/names) make the surface practical for common queries.
Real-time transit stops, routes, arrivals, vehicle positions, and schedules via OneBusAway APIs.
Read-only public transit departures, stop search, and city coverage for bus and train users.
Get real-time NYC bus arrivals, live vehicle locations, and service alerts. Plan trips between any…
MBTA MCP — Boston real-time transit via the MBTA v3 API (api-v3.mbta.com)