Skip to main content
Glama
deenesjakoruzh
RyK57

transit

by RyK57
OverviewSchemaSourceRelated ServersDiscussionsAdmin
TypeScript
Hybrid

transit-mcp-server

Servidor MCP para la API de tránsito 511.org SF Bay Open Data. Proporciona a un LLM datos de tránsito en vivo del Área de la Bahía: agencias, rutas, paradas, salidas en tiempo real, posiciones de vehículos y alertas de servicio, en BART, Muni, AC Transit, Caltrain, VTA y cualquier otro operador que reporte a 511.

6 herramientas, todas de solo lectura.

Requisitos

Related MCP server: mcp-bart

Instalación

npm install
npm run build

Configuración

{
  "mcpServers": {
    "transit": {
      "command": "node",
      "args": ["/absolute/path/to/transit-mcp-server/dist/index.js"],
      "env": { "TRANSIT_511_API_KEY": "your-token-here" }
    }
  }
}

Variable

¿Requerida?

Valor predeterminado

Propósito

TRANSIT_511_API_KEY

sí

—

Token de https://511.org/open-data/token

TRANSIT_511_BASE_URL

no

https://api.511.org

Sobrescribir el host de la API

TRANSIT_511_REQUEST_TIMEOUT_MS

no

30000

Tiempo de espera por solicitud

TRANSPORT

no

stdio

stdio o http

PORT / HOST

no

3000 / 127.0.0.1

Dirección de enlace del transporte HTTP

MCP_PATH_SECRET

cuando se aloja

—

Sirve el endpoint en /mcp/<secreto>. Requerida cuando HOST no es loopback

ALLOWED_ORIGINS

no

localhost + claude.ai

Lista de orígenes permitidos separados por comas

La cuota es la principal limitación

511 permite 60 solicitudes por hora por clave, compartidas entre todos los endpoints. Eso es lo suficientemente bajo como para dar forma a cómo se deben usar estas herramientas:

  • Resuelve los códigos de operador y de parada una vez, y luego reutilízalos. No cambian.

  • Prefiere transit_list_service_alerts sin operator_id: una sola llamada cubre todas las agencias.

  • Nunca hagas polling de transit_next_departures en un bucle. Diez comprobaciones durante un viaje al trabajo son una sexta parte del presupuesto por hora.

transit_list_operators informa cuánto presupuesto queda, leído del encabezado RateLimit-Remaining que 511 devuelve en cada respuesta. Superar la cuota devuelve 429; solicita un aumento a transitdata@511.org.

Despliegue (para conectores de Claude mobile / claude.ai)

Misma forma que cualquier servidor MCP alojado: genera un secreto de ruta con openssl rand -hex 32, establece TRANSIT_511_API_KEY y MCP_PATH_SECRET en el panel de la plataforma, y el Dockerfile y railway.json incluidos funcionan tal cual en Railway, Render o Fly. El servidor se niega a iniciarse en una interfaz pública sin un secreto. /healthz es una sonda de actividad sin autenticación.

Luego, en claude.ai en un navegador: Personalizar → Conectores → Añadir conector personalizado, URL https://tu-app.up.railway.app/mcp/<secreto>.

Herramientas

Red — transit_list_operators, transit_list_lines, transit_find_stops

Tiempo real — transit_next_departures, transit_list_vehicles

Alertas — transit_list_service_alerts

Cada herramienta acepta response_format: "markdown" | "json". Markdown es el valor predeterminado y está optimizado para que un LLM lo lea; JSON es la carga útil estructurada completa. structuredContent siempre se completa independientemente del formato.

Ejemplos

"¿Cuándo es el próximo N Judah?" → transit_find_stops con operator_id="SF", query="judah" para obtener el código de parada, luego transit_next_departures con ese código y line="N".

"¿Está BART funcionando con normalidad?" → transit_list_service_alerts con operator_id="BA".

"¿Algo mal en mi viaje al trabajo?" → transit_list_service_alerts sin operador: una llamada cubre todas las agencias del Área de la Bahía.

"¿Dónde están los trenes ahora mismo?" → transit_list_vehicles con operator_id="BA".

Notas de diseño

Solo lectura por construcción. 511 no publica endpoints de escritura, y cada herramienta lleva readOnlyHint: true. Una prueba lo verifica.

Un operator_id, mapeado por endpoint. 511 llama a este parámetro operator_id en sus endpoints estáticos y agency en los de tiempo real, para el mismo valor. Cada herramienta aquí acepta operator_id y el cliente lo mapea. Esa división es problema de 511, no del llamador.

Los dos endpoints de tiempo real tienen envolturas genuinamente diferentes. StopMonitoring no tiene envoltura raíz Siri; VehicleMonitoring sí la tiene. La especificación publicada muestra una para ambos: la especificación está equivocada, y analizar la forma documentada no devolvería nada en absoluto para las salidas. Ambos se analizan tal como la API en vivo realmente los emite, con una prueba que fija cada uno.

Las llegadas llevan la cuenta regresiva, no las salidas. ExpectedDepartureTime es nulo en prácticamente todas las filas reales, por lo que basar una cuenta regresiva en él mostraría una parada sin servicio. ExpectedArrivalTime es el campo confiable.

Se elimina una marca de orden de bytes UTF-8 antes del análisis. 511 antepone U+FEFF a los cuerpos JSON, lo que hace que un JSON.parse ingenuo falle con una carga útil perfectamente válida. Los fallos de autenticación son texto plano sin BOM, por lo que la eliminación ocurre después de la verificación de estado.

Los valores que parecen números y booleanos a menudo no lo son. Las coordenadas y los rumbos llegan como cadenas JSON, VehicleAtStop es la cadena "false", y "" se usa en todo donde se quiere decir nulo. Coercionar a ciegas convertiría una posición faltante en un 0,0 de apariencia válida frente a la costa de África, por lo que las cadenas vacías se tratan como ausentes en lugar de cero.

El centinela de época cero no es una marca de tiempo. Un viaje que está programado pero no tiene vehículo asignado reporta RecordedAtTime de 1970-01-01T00:00:00Z. Se muestra como "aún no se ha asignado vehículo" en lugar de "registrado hace 56 años".

Los enums de GTFS-Realtime se decodifican. La representación JSON de alertas de 511 emite "effect": 3 donde la representación XML dice SignificantDelays. Tanto la causa como el efecto se mapean de vuelta a palabras.

Las pseudo-agencias internas de 511 se filtran. 5E, 5F, 5O y 5S son 511 Emergency, Flap Sign, Operations y Staff: aparecen en la lista de operadores sin datos de servicio.

Todo es Pacífico. Las marcas de tiempo llegan como UTC y se muestran en America/Los_Angeles, por lo que el cambio de horario de verano se maneja aquí una vez en lugar de que el modelo lo haga dos veces al año. Ten en cuenta que el propio campo TimeZone de 511 reporta America/Vancouver para cada agencia del Área de la Bahía: un error conocido de datos ascendente, ignorado deliberadamente.

La truncación siempre se indica. 511 no pagina; devuelve colecciones completas, y una agencia grande tiene miles de paradas. Las herramientas aceptan un limit del lado del cliente y cada resultado recortado indica cuánto se retuvo, porque una lista acortada silenciosamente se lee como "eso es todo".

Advertencias

  • La cuota por hora es de 60 solicitudes en todos los endpoints. Esta es la restricción vinculante para cualquier flujo de trabajo.

  • Los códigos de operador son fáciles de adivinar mal: VTA es SC (no VT), Capitol Corridor es AM (no CC), Tri Delta es 3D. transit_list_operators imprime estas trampas en su salida.

  • Los códigos de parada pertenecen a un operador y no son intercambiables entre agencias.

  • transit_find_stops filtra en este servidor, por lo que una consulta estrecha no ahorra cuota: la lista completa de paradas se obtiene de todos modos.

  • Las predicciones en tiempo real se extienden aproximadamente 90 minutos hacia adelante, y 511 omite la parada final de solo llegada de una ruta del feed de salidas.

  • tripupdates y vehiclepositions son solo protobuf sin opción JSON, por lo que deliberadamente no se exponen: soportarlos significaría asumir una dependencia de protobuf para datos que los endpoints SIRI ya cubren.

Estructura del proyecto

src/
├── index.ts               # entry point, transport selection
├── constants.ts           # enums, limits, operator-code traps
├── types.ts               # interfaces for every 511 entity
├── services/
│   └── transit-client.ts  # fetch wrapper, auth, BOM stripping, quota tracking, errors
├── schemas/
│   ├── inputs.ts          # Zod input schemas
│   └── outputs.ts         # structuredContent schemas
├── formatters/
│   ├── response.ts        # limiting, truncation, Pacific-time rendering
│   └── entities.ts        # per-entity markdown rendering
└── tools/
    ├── network.ts         # operators, lines, stops
    ├── departures.ts      # real-time arrivals and vehicles
    └── alerts.ts          # service alerts

Pruebas

npm run build
npm test            # 42 checks: handshake, BOM, envelopes, quirks, errors (mocked API)
npm run test:http   # 17 checks: config validation, path-secret gating, method handling, origins

Ambas suites se ejecutan contra un mock local que reproduce deliberadamente las peculiaridades reales de 511: el BOM, la envoltura Siri faltante, booleanos y coordenadas como cadenas, el centinela de época y cuerpos de error en texto plano, porque eso es exactamente lo que un cliente ingenuo hace mal.

Available Tools

6 tools
transit_find_stopsFind Transit StopsA
Read-onlyIdempotent

Find an agency's stops by name, and get the stop codes the real-time tools need.

This is the bridge between "Downtown Berkeley" and the code transit_next_departures wants. Pass a query: a large agency has thousands of stops, and 511 returns all of them in one unpaginated response.

Args:

  • operator_id (string): agency code from transit_list_operators

  • query (string): substring match on the stop name — strongly recommended

  • limit (number): maximum stops to return (default: 25)

  • response_format ('markdown' | 'json'): output format (default: 'markdown')

Returns: { "count": number, "total": number, "truncated": boolean, "stops": [ { "id": string, "Name": string, "Location": { "Latitude": string, "Longitude": string } } ] }

Examples:

  • "When's the next train from Downtown Berkeley?" -> operator_id='BA', query='downtown berkeley', then pass the id to transit_next_departures

  • "Find Muni stops on Judah" -> operator_id='SF', query='judah'

  • Don't use when: you already have the stop code

Error Handling:

  • Stop codes belong to ONE operator and are not interchangeable between agencies

  • Filtering happens on this server, so a narrow query does not save quota — the full list is fetched either way

  • 511 allows 60 requests per hour across ALL endpoints

ParametersJSON Schema
NameRequiredDescriptionDefault
limitNoMaximum results to return (max 200)
queryNoCase-insensitive substring match on the stop name. Strongly recommended — a large agency has thousands of stops
operator_idYesOperator code from transit_list_operators, e.g. 'BA' for BART, 'SF' for Muni
response_formatNoOutput format: 'markdown' for human-readable, 'json' for machine-readablemarkdown

Output Schema

ParametersJSON Schema
NameRequiredDescription
countYes
stopsYes
totalYes
truncatedYes

TDQS

A4.7/5.0
Behavior5/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

Annotations already declare readOnlyHint, openWorldHint, idempotentHint, and destructiveHint=false. The description adds meaningful behavior: server-side filtering (no quota savings), unpaginated response from 511, and the 60-request/hour rate limit. It also clarifies that stop codes are operator-specific—valuable context 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.

Conciseness5/5

Is the description appropriately sized, front-loaded, and free of redundancy?

The description is longer but well-structured with clear sections (Args, Returns, Examples, Error Handling). Each section contributes distinct value: the main purpose is front-loaded, examples are concise, and error handling is pertinent. No fluff or redundancy.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness5/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

The description is fully self-contained: it includes the return object structure, examples mapping to actual operator IDs, error-handling notes, rate limits, and the relationship to sibling tools. With an output schema present and annotations covering safety, nothing an agent needs to invoke correctly is missing.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters3/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Schema description coverage is 100%, so the baseline is 3. The description's Args section largely repeats schema descriptions (e.g., 'operator_id' from transit_list_operators, 'query' substring match and 'strongly recommended'). It adds no new per-parameter semantics; the 'strongly recommended' phrasing already appears in the schema. The unpaginated response note is behavioral, not parameter-specific.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

The description clearly identifies the tool's purpose: 'Find an agency's stops by name, and get the stop codes the real-time tools need.' It explicitly ties the tool to transit_next_departures, distinguishing it from other transitive tools. The purpose is specific and actionable.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines5/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

The description provides explicit usage context: 'This is the bridge between "Downtown Berkeley" and the code transit_next_departures wants.' It also states a clear when-not: 'Don't use when: you already have the stop code.' Examples illustrate typical queries, making the intended conditions unambiguous.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

transit_list_linesList Transit LinesA
Read-onlyIdempotent

List an agency's routes.

Use this to resolve a line name a user mentions into the id the real-time tools filter on, or to answer "what routes does this agency run".

Args:

  • operator_id (string): agency code from transit_list_operators

  • query (string): substring match on the line name or public code

  • limit (number): maximum lines to return (default: 25)

  • response_format ('markdown' | 'json'): output format (default: 'markdown')

Returns: { "count": number, "total": number, "truncated": boolean, "lines": [ { "Id": string, "Name": string, "PublicCode": string, "TransportMode": string, "Monitored": boolean } ] }

Examples:

  • "What BART lines are there?" -> operator_id='BA'

  • "Is there an N line on Muni?" -> operator_id='SF', query='N'

  • Don't use when: you want stops on a line (use transit_find_stops)

Error Handling:

  • A large agency returns many lines; use query to narrow rather than raising limit

  • 511 allows 60 requests per hour across ALL endpoints

ParametersJSON Schema
NameRequiredDescriptionDefault
limitNoMaximum results to return (max 200)
queryNoCase-insensitive substring match on the line name or public code
operator_idYesOperator code from transit_list_operators, e.g. 'BA' for BART, 'SF' for Muni
response_formatNoOutput format: 'markdown' for human-readable, 'json' for machine-readablemarkdown

Output Schema

ParametersJSON Schema
NameRequiredDescription
countYes
linesYes
totalYes
truncatedYes

TDQS

A4.7/5.0
Behavior5/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

The annotations already declare readOnly, idempotent, and non-destructive hints. The description goes further by disclosing the response shape, the truncation flag, the possibility of many results for large agencies, and the 60-request-per-hour rate limit across all endpoints. This adds behavioral context beyond the annotations and contradicts nothing.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness5/5

Is the description appropriately sized, front-loaded, and free of redundancy?

The description is well-organized with clear sections (Summary, Args, Returns, Examples, Error Handling). It front-loads the purpose and usage, and every sentence adds value—examples, exclusions, rate limits, and return format. Despite covering many aspects, it remains concise and skimmable.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness5/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

The tool has 4 parameters, an output schema, and sits among 5 siblings. The description covers all essentials: purpose, parameter examples, return structure, error handling, and even the rate limit. There is nothing an agent needs to know to call this correctly that is missing, and the output schema already documents the return shape.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters3/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

The schema covers all parameters with detailed descriptions (100% coverage), so the description does not need to duplicate that. It adds a few examples (operator_id='BA', 'SF') and clarifies that query is a case-insensitive substring, but these are minor enhancements. The baseline of 3 is appropriate because the schema already carries the main semantic load.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

The description states a specific verb ('List') and resource ('an agency's routes'), and immediately explains its practical use (resolving a line name into an id, answering route questions). It explicitly distinguishes itself from a sibling tool ('Don't use when: you want stops on a line (use transit_find_stops)'), which is a clear differentiation.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines5/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

It gives explicit 'Use this to...' guidance and an explicit 'Don't use when...' with a named alternative. Examples illustrate common queries, and the error handling section advises when to prefer query narrowing over raising limit. This leaves no ambiguity about when to invoke this tool.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

transit_list_operatorsList Transit OperatorsA
Read-onlyIdempotent

List Bay Area transit agencies and their operator codes.

Start here. Every other tool needs an operator code, and this is what produces them — BART is 'BA', Muni is 'SF', AC Transit is 'AC', Caltrain is 'CT'.

The Monitored flag matters: agencies reporting real-time data support live departures and vehicle positions, while schedule-only agencies do not.

Args:

  • monitored_only (boolean): only agencies publishing real-time data (default: false)

  • limit (number): maximum operators to return (default: 50)

  • response_format ('markdown' | 'json'): output format (default: 'markdown')

Returns: { "count": number, "total": number, "truncated": boolean, "operators": [ { "Id": string, "Name": string, "Monitored": boolean, "PrimaryMode": string, "TimeZone": string } ] }

Examples:

  • "What transit agencies are there?" -> call with no arguments

  • "Which ones have live tracking?" -> monitored_only=true

  • Call this first whenever the user names an agency, to resolve its code

Error Handling:

  • Ignore the TimeZone field: 511 reports "America/Vancouver" for every Bay Area agency, which is a known upstream data bug. Everything here is Pacific time

  • 511-internal pseudo-agencies (5E, 5F, 5O, 5S) are filtered out — they carry no service data

  • 511 allows 60 requests per hour across ALL endpoints, so cache this rather than re-fetching

ParametersJSON Schema
NameRequiredDescriptionDefault
limitNoMaximum results to return (max 200)
monitored_onlyNoOnly agencies that publish real-time data, excluding schedule-only ones
response_formatNoOutput format: 'markdown' for human-readable, 'json' for machine-readablemarkdown

Output Schema

ParametersJSON Schema
NameRequiredDescription
countYes
totalYes
operatorsYes
truncatedYes

TDQS

A4.9/5.0
Behavior5/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

Annotations already declare readOnlyHint=true, openWorldHint=true, idempotentHint=true, and destructiveHint=false. The description adds substantial behavioral context beyond those: the Monitored flag semantics, the known TimeZone bug (inaccurate 'America/Vancouver' values), the filtering of pseudo-agencies (5E, 5F, 5O, 5S), and the rate limit (60 req/hour across all endpoints) with caching advice. This is exactly the kind of operational detail an agent needs and the annotations do not provide.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness5/5

Is the description appropriately sized, front-loaded, and free of redundancy?

The description is thorough but efficiently organized into clear sections (Args, Returns, Examples, Error Handling). Every sentence adds value: the opening line is the purpose, the 'Start here' directive is front-loaded, examples are concise, and error handling covers real-world quirks. No redundancy or filler—each element earns its place.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness5/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

For a tool with 3 optional parameters and an output schema, the description covers everything needed to call it correctly: it provides default values, return shape (matching the output schema), examples, and critical edge cases (TimeZone bug, filtering, rate limit). Even though the output schema exists, the description's inclusion of the return shape is redundant but not harmful; the error-handling notes are essential and not available anywhere else. The description is fully sufficient for an agent to use this tool without surprises.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters4/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Schema coverage is 100%, so baseline is 3. The description goes beyond the schema by explaining the semantic impact of monitored_only (supports live departures/vehicle positions vs schedule-only), giving concrete usage examples for each parameter, and clarifying the response_format default. It also mentions the default for limit in prose. While the schema already has descriptions, the description adds contextual meaning that helps an agent decide parameter values appropriately.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

The description opens with a clear verb-resource pair: "List Bay Area transit agencies and their operator codes." It immediately differentiates from siblings by noting "Every other tool needs an operator code, and this is what produces them," which is unique among the sibling tools. The specific examples (BART='BA', Muni='SF') further cement the purpose.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines5/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

Explicitly instructs when to use: "Start here." and "Call this first whenever the user names an agency, to resolve its code." It also explains the monitored_only flag's relevance for choosing between live vs schedule-only agencies, effectively guiding the agent on when to apply different parameter values. No alternatives are named, but the tool is clearly positioned as the entry point, so exclusion guidance is unnecessary.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

transit_list_service_alertsList Transit Service AlertsA
Read-onlyIdempotent

Service alerts — delays, outages, detours and planned disruptions.

Answers "is BART running normally", "why is my line delayed" and "anything I should know before I leave". Omit operator_id to sweep every Bay Area agency in a single request, which is also the kinder option against the hourly quota.

Args:

  • operator_id (string): limit to one agency; omit for all operators

  • query (string): substring match on the headline or description

  • active_only (boolean): only alerts whose active period covers now (default: true)

  • limit (number): maximum alerts to return (default: 25)

  • response_format ('markdown' | 'json'): output format (default: 'markdown')

Returns: { "count": number, "total": number, "truncated": boolean, "alerts": [ { "id": string, "header": string, "description": string, "effect": string, "cause": string, "routes": [string], "start": number, "end": number } ] // epoch SECONDS }

Examples:

  • "Any BART delays?" -> operator_id='BA'

  • "Anything wrong on my commute?" -> omit operator_id, read across agencies

  • "Weekend track work?" -> active_only=false, query='weekend'

  • Don't use when: you want a specific stop's arrivals (use transit_next_departures)

Error Handling:

  • No alerts is genuinely good news, not an error — it means normal service

  • Timestamps here are epoch seconds, unlike the ISO strings elsewhere in this API

  • Setting active_only=false surfaces future planned work as well as current problems

  • 511 allows 60 requests per hour across ALL endpoints

ParametersJSON Schema
NameRequiredDescriptionDefault
limitNoMaximum results to return (max 200)
queryNoCase-insensitive substring match on the alert headline or description
active_onlyNoOnly alerts whose active period covers now
operator_idNoLimit to one agency; omit for alerts across every Bay Area operator
response_formatNoOutput format: 'markdown' for human-readable, 'json' for machine-readablemarkdown

Output Schema

ParametersJSON Schema
NameRequiredDescription
countYes
totalYes
alertsYes
truncatedYes

TDQS

A5/5.0
Behavior5/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

Annotations only indicate read-only, idempotent, and non-destructive behavior. The description adds crucial behavioral nuance: no alerts means normal service (not an error), timestamps are epoch seconds unlike the ISO strings in other API endpoints, active_only=false surfaces future planned work, and rate limits. This goes well 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.

Conciseness5/5

Is the description appropriately sized, front-loaded, and free of redundancy?

Structured with Args, Returns, Examples, and Error Handling sections. The purpose and primary use cases are front-loaded in the first sentence, and every sentence adds value—examples, rate limits, or format distinctions. No redundancy or fluff.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness5/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

The output schema is present and the description includes a return format example with field types. It covers error handling (no alerts ≠ error), rate limits, timestamp format, and sibling routing. For a 5-parameter tool with no required parameters, this is fully complete for an agent.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters5/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Schema coverage is 100%, so the baseline is 3, but the description adds substantial meaning: operator_id omission sweeps all agencies and is 'kinder' against quota, active_only=false is tied to 'weekend track work' examples, and query is illustrated with a practical use case. These contextual insights help an agent choose parameter values effectively.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

The description states a specific verb ('List') and resource ('Transit Service Alerts'), and grounds it with concrete questions it answers ('is BART running normally', 'why is my line delayed'). It explicitly differentiates from the sibling transit_next_departures by naming when not to use it, making the purpose unmistakable.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines5/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

Explicitly states when to use the tool (answering delay/outage questions) and when not to ('Don't use when: you want a specific stop's arrivals (use transit_next_departures)'). Also gives operational guidance, such as omitting operator_id to sweep all agencies and conserve quota—practical context an agent needs.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

transit_list_vehiclesList Transit Vehicle PositionsA
Read-onlyIdempotent

Live positions of an agency's vehicles currently in service.

Answers "where are the trains right now" and "how many buses are running on this line". For "when does one get to me", use transit_next_departures instead.

Args:

  • operator_id (string): agency code from transit_list_operators

  • line (string): only vehicles on this line, matched on name or id

  • limit (number): maximum vehicles to return (default: 25)

  • response_format ('markdown' | 'json'): output format (default: 'markdown')

Returns: { "count": number, "total": number, "truncated": boolean, "vehicles": [ { "line": string, "vehicle": string, "destination": string, "latitude": number | null, "longitude": number | null, "bearing": number | null, "recorded_at": string } ] }

Examples:

  • "How many Muni trains are running on the N?" -> operator_id='SF', line='N'

  • "Where are the BART trains?" -> operator_id='BA'

  • Don't use when: you want arrival times at a stop (use transit_next_departures)

Error Handling:

  • Coordinates arrive as strings and may be empty; those vehicles report "position unavailable" rather than a false 0,0

  • An agency with no real-time feed returns nothing — check Monitored in transit_list_operators

  • 511 allows 60 requests per hour across ALL endpoints

ParametersJSON Schema
NameRequiredDescriptionDefault
lineNoOnly vehicles on this line
limitNoMaximum results to return (max 200)
operator_idYesOperator code from transit_list_operators, e.g. 'BA' for BART, 'SF' for Muni
response_formatNoOutput format: 'markdown' for human-readable, 'json' for machine-readablemarkdown

Output Schema

ParametersJSON Schema
NameRequiredDescription
countYes
totalYes
vehiclesYes
truncatedYes

TDQS

A4.8/5.0
Behavior5/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

Annotations already signal read-only, idempotent, and open-world behavior. The description adds valuable, non-obvious behavioral details: coordinates arrive as strings and may be empty, agencies with no real-time feed return nothing (with a pointer to check 'Monitored'), and a 60-requests-per-hour rate limit. This goes well beyond the annotations and prepares the agent for realistic edge cases.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness4/5

Is the description appropriately sized, front-loaded, and free of redundancy?

The description is longer than strictly necessary because it repeats the Args and Returns sections that already exist in the schema. However, it is well-structured with clear sections (purpose, args, returns, examples, error handling) and the most important guidance (purpose and alternative) is front-loaded. Every section adds unique value (examples, error handling, rate limits), so the length is justified despite some redundancy.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness5/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

For a list tool with an output schema and read-only annotations, the description covers all essential context: proper use cases, alternatives, parameter semantics, edge cases (missing coordinates, empty feeds, rate limits), and concrete examples. Nothing an agent needs to call it correctly is missing.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters4/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Schema coverage is 100%, so the schema fully documents each parameter. The description adds some meaning beyond that: it clarifies that 'line' matches on name or id, and provides concrete examples (operator_id='SF', line='N') that illustrate how parameters map to real queries. Defaults for limit and response_format are repeated, but the examples and matching detail provide modest added value.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

The description opens with a clear, specific statement of what the tool does: 'Live positions of an agency's vehicles currently in service.' It names the resource (vehicles) and action (list positions), and explicitly differentiates from a sibling tool ('For "when does one get to me", use transit_next_departures instead'), so an agent can distinguish it without opening schemas.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines5/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

Provides explicit when-to-use guidance: answers 'where are the trains right now' and 'how many buses are running on this line', and gives a direct exclusion: 'Don't use when: you want arrival times at a stop' with the alternative named. Examples reinforce correct usage and operator/line selection.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

transit_next_departuresNext Transit DeparturesA
Read-onlyIdempotent

Live arrival predictions for the next vehicles at a stop.

This is the tool for "when is my next train/bus". It returns real-time predictions, not the printed timetable, and reports minutes-from-now alongside Pacific clock time.

Args:

  • operator_id (string): agency code from transit_list_operators, e.g. 'BA', 'SF'

  • stop_code (string): stop code from transit_find_stops. Codes belong to ONE operator and are not interchangeable between agencies

  • line (string): only departures on this line, matched on name or id

  • limit (number): maximum departures to return (default: 10)

  • response_format ('markdown' | 'json'): output format (default: 'markdown')

Returns: { "count": number, "total": number, "truncated": boolean, "operator": string, "stop_code": string, "stop_name": string, "retrieved_at": string, "departures": [ { "line": string, "destination": string, "expected": string, "aimed": string, "minutes": number, "vehicle": string | null, "at_stop": boolean } ] }

Examples:

  • "When's the next N Judah?" -> operator_id='SF', stop_code from transit_find_stops, line='N'

  • "Next BART from Downtown Berkeley?" -> operator_id='BA', the stop's code

  • Don't use when: you want the scheduled timetable rather than live predictions

Error Handling:

  • An empty result usually means service has ended for the night, or the stop is a route's final stop — 511 omits arrival-only terminals from this feed

  • Predictions extend roughly 90 minutes ahead; nothing beyond that appears

  • Rows reading "scheduled only, no live prediction" have no vehicle assigned yet

  • 511 allows 60 requests per hour across ALL endpoints — never poll this in a loop

ParametersJSON Schema
NameRequiredDescriptionDefault
lineNoOnly departures on this line, matched against the line name or id
limitNoMaximum results to return (max 200)
stop_codeYesStop code from transit_find_stops. Codes are specific to one operator
operator_idYesOperator code from transit_list_operators, e.g. 'BA' for BART, 'SF' for Muni
response_formatNoOutput format: 'markdown' for human-readable, 'json' for machine-readablemarkdown

Output Schema

ParametersJSON Schema
NameRequiredDescription
countYes
totalYes
operatorNo
stop_codeNo
stop_nameNo
truncatedYes
departuresYes
retrieved_atNo

TDQS

A5/5.0
Behavior5/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

Annotations already declare readOnly, idempotent, and non-destructive hints. The description adds substantial behavioral context: real-time vs static, minutes-from-now plus Pacific clock time, 'scheduled only' rows, a 90-minute prediction horizon, empty results meaning (end of service or arrival-only terminal), and a 60-requests-per-hour rate limit. This goes far beyond the annotations, providing critical operational details an agent needs to interpret results and avoid misuse.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness5/5

Is the description appropriately sized, front-loaded, and free of redundancy?

Structured with clear sections (Purpose, Args, Returns, Examples, Error Handling) and front-loaded with the core purpose. Every sentence adds value; error handling is critical for correct agent behavior. It is detailed but not flabby — the length is justified by the operational caveats and edge cases.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness5/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

For a tool with 5 parameters and an output schema, the description covers all essential usage aspects: parameter semantics, error scenarios (empty results, truncated predictions), time horizon, rate limits, and a full output structure in the Returns section. Even though an output schema is provided, the description's explicit sample JSON and explanation of fields further aid the agent. Nothing needed for correct invocation is missing.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters5/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Even though schema coverage is 100%, the description enriches every parameter: it provides real examples (operator_id 'BA'/'SF', stop_code from transit_find_stops), warns that stop codes are operator-specific and not interchangeable, clarifies line matching ('matched on name or id'), and explains response_format options. The Args section adds value beyond the schema's terse descriptions, especially for stop_code and line.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

Description states a specific verb and resource: 'Live arrival predictions for the next vehicles at a stop' and explicitly positions it as 'the tool for when is my next train/bus'. It distinguishes from the printed timetable and clarifies scope (real-time, not static), making it clearly different from sibling tools that list operators, stops, lines, vehicles, or alerts.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines5/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

Provides explicit when-to-use guidance via examples ('When's the next N Judah?' maps to specific parameters), an explicit 'Don't use when' exclusion for scheduled timetable, and references prerequisite tools (transit_list_operators, transit_find_stops) for parameter sourcing. It also includes rate-limit guidance ('never poll this in a loop'), which is directly relevant to usage.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

  1. 6 tool updatesv1.0.0
    • First observedtransit_find_stops
    • First observedtransit_list_lines
    • First observedtransit_list_operators
    • First observedtransit_list_service_alerts
    • First observedtransit_list_vehicles
    • First observedtransit_next_departures
Deploy Server
One endpoint for every MCP toolRayrun runs every MCP server behind one URL and holds the credentials. Each call is decided per client, tool, and argument, then recorded.ray.runAd

TDQS

A4.8/5.0

Scored across 6 tools

Disambiguation5/5

Each tool targets a distinct resource: operators, lines, stops, departures, vehicles, and alerts. Descriptions explicitly cross-reference when to use which (e.g., 'Don't use when' sections), eliminating ambiguity. No two tools overlap in purpose.

Naming Consistency5/5

All tools share the 'transit_' prefix and follow a consistent verb-object pattern: list_operators, list_lines, find_stops, next_departures, list_vehicles, list_service_alerts. Minor variation between 'list', 'find', and 'next' but the style is uniform and predictable.

Tool Count5/5

Six tools are well-scoped for a transit information server, covering discovery (operators, lines, stops), real-time queries (departures, vehicles), and alerts. No tool is redundant, and the count feels neither sparse nor bloated.

Completeness4/5

The tool surface covers the full workflow of querying transit info: discover agencies, lines, stops, then get live departures, vehicle positions, and alerts. A minor gap is the lack of a dedicated scheduled timetable tool, though the departures endpoint includes aimed times. Overall, the domain is well covered.

Maintenance

ActivitySlowing
ResponsivenessNo issues

Resources

Related MCP Connectors

Related MCP Servers