Loading ToolBench
Search USGS and EMSC seismic data — real-time feeds, event queries, and earthquake counts via MCP. STDIO or Streamable HTTP.
Strong tool definitions with comprehensive descriptions and well-structured schemas. All 4 tools have detailed, LLM-optimized descriptions (200-500 chars) explaining WHAT, WHEN, and HOW. Input schemas are complete with types and descriptions for all parameters. Tool names follow verb_noun convention (earthquake_*). Output schemas are documented in descriptions. Main gaps: no explicit error handling guidance in descriptions, no per-tool security/permission declarations, and output field mappings could be more explicit for chaining.
Count earthquakes matching filters without fetching full records. Use for statistical queries ("how many M5+ earthquakes in 2025?") or to gauge result size before calling earthquake_search. Omitting start_time counts only the last 30 days, so pass an explicit range for any period-specific question; queryEcho reports the window and filters the count actually covers. When exceeds_limit is true, the count exceeds 20,000 and a full search would be truncated — narrow filters before fetching. USGS returns the max_allowed cap (20,000); EMSC count endpoint does not return this field (max_allowed will be null). Counts can be scoped to a rectangular study area with min_latitude, max_latitude, min_longitude, and max_longitude — each independently optional. Combining the box with the lat/lon/radius circle intersects the two, counting only events inside both. Both catalogs include non-tectonic records, so a radius over a mining region counts quarry blasts alongside earthquakes — pass event_type="earthquake" on USGS to exclude them. USGS-specific filters (alert_level, event_type, min_felt, min_significance) are not sent when source=emsc — the response names them in ignoredFilters.
Fetch detail for a specific earthquake by USGS event ID. Returns the same normalized event a search result carries, plus a projection of the analysis products only the single-event response holds: PAGER impact alert and report link, ShakeMap peak intensity and ground motion, DYFI felt-report totals, the moment-tensor focal mechanism, landslide and liquefaction alerts, origin quality (azimuthal gap, station count, location uncertainty), and finite-fault rupture dimensions. Products are omitted when USGS produced none — a small automatic event typically has no detail at all, while a large reviewed one has most of it. Event IDs appear in the "id" field of earthquake_get_feed and earthquake_search results. This tool is USGS-only — EMSC events have no per-event detail endpoint.
Fetch a USGS pre-computed real-time earthquake feed by magnitude tier and time window. These feeds are CDN-cached by USGS and faster and more available than the query API — use them for "what's happening now" queries. "all" includes microseisms (M<1); "significant" is a USGS curation based on magnitude, felt reports, and PAGER impact estimates. "hour" returns 0–10 events typically; "month" can exceed 10,000 for the "all" tier, so results are returned a page at a time: count is the page size, totalCount the whole feed, and nextCursor the input for the following page. The cursor is opaque and must be passed back verbatim — unlike earthquake_search, these feeds have no upstream paging parameter and USGS regenerates them about once a minute, so a numeric offset across two calls would skip or repeat events. For historical or filtered queries, use earthquake_search instead.
No explicit error recovery guidance in tool descriptions. Descriptions explain what tools do but not what to do if they fail (e.g., 'If count exceeds 20,000, narrow filters' is present, but no guidance for API failures, timeouts, or invalid parameters).
Output schemas not formally documented in JSON Schema format. Descriptions mention returned fields (e.g., 'queryEcho', 'exceeds_limit', 'nextCursor') but no structured output schema is visible in tool definitions.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint) visible in definitions. All tools are read-only, which should be declared via toolAnnotations for clarity.
Inferred effective spec: 2026-07-28+.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 67 | 2026-07-28+ | v2 |
Search earthquakes by time range, magnitude, depth, location radius, PAGER alert level, or felt reports. Supports USGS (global, richer metadata: PAGER, DYFI, ShakeMap) and EMSC, an independent global catalog operated by the European-Mediterranean Seismological Centre. For location-based queries, provide latitude, longitude, and radius_km together. A rectangular study area is expressed with min_latitude, max_latitude, min_longitude, and max_longitude — each independently optional, so a single edge is a valid constraint. Combining the box with the lat/lon/radius circle intersects the two, returning only events inside both. Both catalogs include non-tectonic records (quarry blasts, explosions) — every event carries its event_type, and event_type="earthquake" filters the rest out on USGS. USGS-specific filters (alert_level, event_type, min_felt, min_significance) are not sent when source=emsc — the response names them in ignoredFilters. Use earthquake_count first to gauge result size before requesting large result sets. A single call returns at most 20,000 events; larger result sets are retrieved by paging with offset, which is passed straight through to the upstream FDSN API. When a result is capped, nextOffset carries the offset for the following page and totalCount the full match count — or countUnavailable reports that the count lookup failed and the total is unknown.
Parameter relationships not fully documented. earthquake_search and earthquake_count accept both radius-based (lat/lon/radius_km) and bounding-box (min/max_latitude/longitude) filters; the intersection logic is explained but mutual exclusivity/combination rules could be clearer.
No pagination limit enforcement documented. earthquake_get_feed mentions 'max 1000' for limit, but earthquake_search mentions 'max 20000', no guidance on recommended limits to avoid context window exhaustion.