MCP server for CATA bus system live and static schedule data
The server demonstrates solid baseline implementation with clear, well-structured tool definitions and functional schemas across all 7 tools. Most tools have appropriate verb-noun naming (list_routes_tool, search_stops_tool, next_arrivals_tool, vehicle_positions_tool, trip_alerts_tool). Descriptions are present and informative (ranging 80 - 250 chars). Input schemas are properly typed with JSON Schema. However, the server has several material gaps: (1) output schemas are not documented, responses are described in prose ('returns list of...') but no structured output schema is provided to the LLM, (2) three tools (health_check, initialize_data, list_routes_tool) have empty or trivial input schemas but their descriptions lack context on when/why to call them, (3) error handling guidance is minimal, no recovery hints or error classification, (4) tool naming inconsistency, 'list_routes_tool' and 'search_stops_tool' use '_tool' suffix which is redundant and non-standard, (5) no pagination/limit parameters despite tools returning lists (search_stops, next_arrivals, vehicle_positions, trip_alerts could return 50+ items), (6) no per-tool annotations (readOnlyHint, idempotentHint) visible, indicating missing protocol features. Composition is solid: each tool has one responsibility, and natural-language parameters (query string, stop_id, route_id) match user intents. Descriptions clearly explain what each tool does and when to use it. Security is adequate for read-only tools. Response field naming is generally consistent (using 'stop_id', 'route_id' in both search and query tools). The server is suitable for production use but would benefit from output schema documentation, pagination support, and error handling enrichment.
Ultra-fast health check optimized for cloud pre-flight validation.
Manually trigger GTFS data initialization.
List all available bus routes. Returns a list of routes with their ID, short name, long name, and color.
Get next arrivals at a specific stop. Args: stop_id: The stop ID to query horizon_minutes: How many minutes ahead to look (default 30) Returns: List of upcoming arrivals with trip ID, route ID, arrival time, and delay
Search for stops by name or ID. Args: query: Search query string to match against stop names, IDs, or descriptions Returns: List of matching stops with their ID, name, latitude, and longitude
Get current service alerts. Args: route_id: Optional route ID to filter alerts (if None, returns all) Returns: List of alerts with route ID, header, description, and severity
No structured output schemas documented in tool definitions. All tools describe their responses in prose ('Returns list of...') but LLMs need JSON Schema output specs to plan downstream calls and extract fields safely.
Tool names use redundant '_tool' suffix (list_routes_tool, search_stops_tool, etc.). Canonical naming is verb_noun without suffix. This adds 5 chars and no semantic value, increasing name collisions.
No pagination, limit, or offset parameters on list-returning tools. search_stops, next_arrivals, vehicle_positions, and trip_alerts could return 50+ items. Unbounded results blow context windows and waste tokens. Tools need page_size/offset or cursor parameters.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 54 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 0 | - | v1 |
Get current positions of vehicles on a specific route. Args: route_id: The route ID to filter by (e.g., "BL" for Blue Loop) Returns: List of vehicle positions with ID, coordinates, bearing, and speed
Minimal error handling guidance. No recovery hints (e.g., 'stop_id not found, call search_stops() first'), no error classification (retryable vs user-fixable), and no actionable error messages. Agents cannot self-correct or plan recovery.
health_check and initialize_data have insufficient descriptions (<50 chars, failing baseline). health_check lacks context on 'cloud pre-flight validation'. initialize_data lacks side-effect documentation and clear trigger conditions.
No tool annotations visible in schema (readOnlyHint, idempotentHint, destructiveHint). All 7 tools are read-only operations except initialize_data (which mutates state), but annotations are missing. This prevents protocol-aware clients from optimizing or restricting tool access.