MCP server for Wildfire Detection using NASA FIRMS data, weather forecasting, air pollution monitoring, and fire fuel analysis
The server defines 6 tools with present descriptions and visible schema structures. However, schemas are incomplete (missing type information in some nested objects), descriptions vary significantly in quality (some are verbose and contain API documentation noise), and there is NO error handling guidance or recovery patterns documented. Tool naming is reasonable but conventional. Parameters lack enumeration constraints where applicable (e.g., date format). The `bounding_box` parameter is typed as an unstructured object with no field documentation. Output schemas are not formally documented, responses are converted to 'human readable' strings, which defeats structured agent reasoning. No tool annotations present (readOnlyHint, destructiveHint, idempotentHint). Average across tools is 58, reflecting fair quality with significant room for improvement.
Get the address for the given latitude and longitude.
Fetch current air pollution data for the given latitude and longitude.
Air Pollution API provides historical air pollution data for any coordinates on the globe.
Get potential wildfires in Iran for a specific date using NASA FIRMS data. This function is a starting point for further analysis of detected fires.
You can search weather forecast for 5 days with data every 3 hours by geographic coordinates. All weather data can be obtained in JSON and XML formats.
Check if the given latitude and longitude are in a potential wildfire fuel area.
Output schemas not documented. All tools convert responses to unstructured 'human readable' strings via `.to_human_readable()`. LLMs cannot parse structured fields, cannot chain results to downstream tools, and must interpret freetext to extract coordinates, fire IDs, or other identifiers for follow-up calls. This breaks tool composition (pattern:tool-chain). Example: get_potential_wildfires returns a formatted string instead of a list of {latitude, longitude, fire_id, frp, brightness} objects, forcing manual parsing.
bounding_box parameter is typed as object with no field schema. Users cannot infer what fields it expects (min_lat? max_lat? south? north?).
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 45 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 0 | - | v1 |
No error handling or recovery guidance. Tools return no documented error cases or recovery hints. If NASA FIRMS API fails, if coordinates are out of range, if date format is invalid, or if API credentials are missing, the agent has no guidance on what to do next.
Parameter descriptions contain API documentation boilerplate that wastes tokens and confuses LLMs. Example: 'Latitude. If you need the geocoder to automatic convert city names and zip-codes to geo coordinates and the other way around, please use our Geocoding API', this is 153 characters of irrelevant API promotion. Affects: get_weather_forecast, get_current_air_pollution_data, get_historical_air_pollution_data.
No enumeration constraints on string parameters. date parameter accepts 'YYYY-MM-DD' format with no regex pattern, no enum validation. If an LLM passes '2024-13-45' or 'January 15', the tool fails with no guidance on correction.
No numeric parameter bounds documented. frp (Fire Radiative Power) defaults to 1 and bright_ti4 defaults to 50, but acceptable ranges are not stated. Can LLMs pass frp=99999? bright_ti4=-100?
Tool descriptions are generic. get_address is 'Get the address for the given latitude and longitude.', missing WHEN to use this tool vs. other address lookups, what format the output is in, whether it geocodes or reverse-geocodes, or what geographic region it covers. When should the LLM call it instead of a similar tool? What does it return?'
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). All six tools are read-only (they query external APIs and make no mutations), but this is not declared. Agents cannot know which tools are safe to retry and which have side effects.