DisasterAWARE's Historical Hazards MCP server connects the AI assistant you already use to DisasterAWARE's historical hazard archive. You ask a question in plain language: "Which flood rose highest during Hurricane Harvey, and at which gauge?" The assistant runs the searches, counts, rankings and lookups against the archive and explains the answer, including where each number comes from.
There is no query language to learn and no GIS to run. The assistant works out which filters, places, dates and measures your question needs, and you can keep asking follow-ups until you have what you came for.
MCP, the Model Context Protocol, is an open standard for connecting AI assistants to outside data and tools. An MCP server tells the assistant what it can look up and how; the assistant decides which lookups a question needs, runs them and reads the results before it answers.
The Historical Hazards MCP server gives an assistant a set of purpose-built tools over the archive: one that describes what the archive contains, one that lists the measurements available for each hazard type, one that finds and sorts events by any filter or measurement, one for counts, totals, averages, percentiles and rankings, one that builds a hazard history for a place, and two that open a single event and the evidence behind its numbers. You never call these yourself. Knowing they exist simply helps you ask better questions.
The archive holds nearly 680,000 historical hazard records across about 30 hazard types: earthquakes, floods, wildfires, tropical cyclones, tornadoes, severe storms, winter storms, extreme temperatures, volcanoes, tsunamis, landslides, smoke and more.
Hundreds of thousands of those records are enhanced with what was measured and reported after the event, from NOAA, the National Weather Service, USGS, FEMA, NIFC and other agencies, satellites and monitoring networks:
Nearly every tornado record and about two thirds of flood and severe thunderstorm records are enhanced, every wildfire record states what kind of evidence it rests on, and modeled population and economic exposure is available on about 260,000 records. Coverage differs by hazard: tornado, hail, wind and flood enhancements come from US agency data, while earthquake and tropical cyclone enhancements are worldwide (cyclone station reports cover the United States and its territories). For the hazard-by-hazard detail, see What's Inside DisasterAWARE's Enhanced Historical Hazard Data.
Every answer below was produced from the archive and checked against the DisasterAWARE Historical Hazards API in October 2026. The archive is rebuilt periodically and enhancement continues after each event, so later answers can differ slightly.
Q: Everything that happened in Texas during Hurricane Harvey, August 25 to September 5, 2017, by hazard type.
A: The archive holds 299 flood records, 63 tornado records, 16 marine warning records and one smoke event that began in Texas in that window. Each flood and tornado record is an alert, such as a flash flood warning or a tornado warning, so the counts describe how the event was warned rather than a count of separate floods or tornadoes.
Why it matters: emergency managers and analysts get the full shape of a past disaster in seconds, as a starting point for after-action reviews and comparisons with the next event.
Q: During Harvey, which Texas flood rose highest above flood stage, and at which gauge?
A: The San Bernard River near Boling. The gauge crested at 43.79 feet late on August 31, 2017 (local time), 25.79 feet above its 18-foot flood stage, in the Major flood category. It is the highest crest in the NWS crest history for that gauge, above the previous record of 42.61 feet from October 1998.
Why it matters: insurance and risk teams can tie a loss to a specific measured peak at a specific place, and see how it ranks against that gauge's own history.
Q: How many people did the Greenfield, Iowa, tornado of May 21, 2024, kill, and where does that figure come from?
A: Five deaths and 35 injuries, according to the NWS damage survey, which maps one continuous EF4 path 43 miles long and up to 1,600 yards wide. NOAA's Storm Events database splits the same tornado into four county segments. Two are short entries in Page and Taylor counties, under two miles together. The other two carry the toll: an EF3 segment in Adams County (24.8 miles, one death) and an EF4 segment in Adair County, which includes Greenfield (15.7 miles, four deaths, 35 injuries and $30 million in property damage). The sources agree on the toll and slice the path differently, and the archive record carries both.
Why it matters: underwriters and analysts can see which source a casualty, rating or damage figure comes from before they compare events or attribute losses by county.
Q: Hurricane Lee did not make a tropical landfall in the United States in 2023. What did the NWS post-storm station network record?
A: Reports from three NWS forecast offices in New England list a highest gust of 58 knots (67 mph) at Matinicus Rock, Maine; a water level 1.62 feet above the normal high-tide line at Chatham, Massachusetts; 6.5 inches of rain near Steuben, Maine; and a lowest measured station pressure of 981.9 mb at Plum Island, Massachusetts (Eastport, Maine, logged an estimated 981.3 mb).
Why it matters: supply chain and facilities teams see what a passing storm actually did at the coast and inland, even when the landfall headline was elsewhere.
Q: Which states had the most tornado alert records in 2024?
A: Texas with 272, Florida 200, Oklahoma 170, Missouri 154 and Illinois 141. These are tornado records, mostly NWS tornado warnings and the confirmed touchdowns under them, so one tornado can appear on several records and a warning can be issued for a tornado that never touched down.
Why it matters: corporate security and site-selection teams get a quick read on where tornado warnings disrupt operations most often, without mistaking alerts for tornadoes.
Put decades of hazard history in your AI assistant
The Historical Hazards MCP server answers questions about floods, storms, fires, earthquakes and cyclones from DisasterAWARE's enhanced archive, with the source behind every number.
Q: Rank the top states by FEMA flood insurance building payments for 2017 floods.
A: Texas, about $6.86 billion; Florida, about $615 million; Louisiana, about $77 million, counting the payments the archive ties to each state's 2017 flood records. These totals are de-duplicated: a flood insurance claim that falls under several overlapping flood warnings is counted once. Adding up every warning's claims without that step would give Texas more than $339 billion, roughly fifty times the real figure. Payments the archive cannot tie to a flood record are left out, so these are not FEMA's own state totals.
Why it matters: insurance and public-sector analysts get loss rankings they can rely on, and a clear demonstration of why counting overlapping alerts naively goes wrong.
Q: What was the worst smoke event by air quality in California in 2020, and which monitoring station recorded it?
A: Seven California smoke events in 2020 reached the top of the US Air Quality Index scale (500, Hazardous) on preliminary EPA AirNow readings. The highest daily (24-hour) fine-particle reading at a California monitor was 487.4 µg/m³ of PM2.5, at Mammoth Lakes on October 17, 2020, in smoke observed from October 14 to 22.
Why it matters: health, facilities and business-continuity planners can find the worst air-quality days for a region and the monitor behind each one.
Q: How many hazard records of each type fall within 50 km of the Port of Houston since 2014, and in which months do floods peak?
A: Floods dominate, with 733 flood records since 2014, concentrated in August and May, followed by September. There are 70 tornado records, 27 of them in 2017, the year of Harvey, and 72 severe thunderstorm records, most of them in May.
Tip: tornado records begin in 2014, and record counts reflect how often warnings were issued, so a window starting in 2014 compares like with like across hazard types.
Why it matters: site-selection, logistics and insurance teams get a defensible first look at a facility's hazard history and seasonality before they commission deeper work.
A few habits get the best answers:
The MCP server is built for answering questions. To bring the records themselves into your own systems or analysis pipelines, use the Historical Hazards API, which serves the same archive with every enhancement.
The documentation has step-by-step setup for Claude, Claude Code, ChatGPT and Cursor, the full list of example questions, and tips for accurate answers.
Explore the historical record
Decades of reconstructed, verified catastrophes — browse the Historical Event Intelligence overview, or request sample data for your own portfolio.
What is the DisasterAWARE Historical Hazards MCP server?
It is a Model Context Protocol server that connects AI assistants to DisasterAWARE's historical hazard archive of nearly 680,000 records across about 30 hazard types. You ask questions in plain language and the assistant runs the searches, counts and rankings for you, and explains where each number comes from.
Which AI assistants can I use with it?
The documentation has setup steps for Claude (claude.ai and the Claude desktop app), Claude Code, ChatGPT and Cursor. Other assistants that support remote MCP servers with OAuth sign-in can also connect. In Claude, custom connectors require a plan that supports them.
What do I need to get access?
A DisasterAWARE account with the MCP access permission added to it. The permission is separate from DisasterAWARE web or API access; your DisasterAWARE administrator or PDC account representative can add it. You then sign in through a DisasterAWARE sign-in page in your browser.
Are tornado counts a count of tornadoes?
No. Tornado records are alerts, mostly NWS tornado warnings and the confirmed touchdowns under them, so one tornado can appear on several records. Ask for counts as tornado records. Totals of deaths, damage and insurance payments are de-duplicated so that each item of evidence is counted once.
Can I use the MCP server to load data into my own systems?
The MCP server is built for answering questions. To bring historical records into your own systems or pipelines, use the DisasterAWARE Historical Hazards API, which serves the same archive with every enhancement.
Sources: DisasterAWARE Historical Hazards API, queried 2026-10-15; DisasterAWARE Historical Hazards MCP documentation. Example sources as named in each answer: NWS (warnings, Damage Assessment Toolkit surveys, post-tropical cyclone reports, river crest history), NOAA NCEI Storm Events, USGS stream gauges, FEMA National Flood Insurance Program claims, EPA AirNow and its partner federal, state, local and tribal air quality agencies (preliminary monitor data). This product uses the Federal Emergency Management Agency's OpenFEMA API, but is not endorsed by FEMA.
Decision-ready intelligence for understanding what's happening, what is impacted, and what to do next. Global hazard, severe weather, catastrophe, and physical risk intelligence trusted by governments, enterprises, insurers, humanitarian organizations, and technology platforms worldwide.
Request DemoContact Us
