Uap Pulse MCP Server

🛸 uap-pulse

An MCP server (and a live map) for the Pentagon's declassified UAP files.

On May 8, 2026, the U.S. Department of War published the first PURSUE Release on war.gov/UFO — 238 declassified UAP/UFO records spanning 1944–2026, across seven federal agencies (War, FBI, NASA, State, CIA, ODNI, Energy). uap-pulse puts that release at your AI agent's fingertips — and on a map.

Government records only. This project uses only public-domain U.S. federal records (17 U.S.C. § 105). No scraped third-party databases, no civilian data behind a ToS. Not affiliated with or endorsed by the U.S. government.

🗺️ Live map: https://uap-pulse.vercel.app · 📦 npm: uap-pulse

---

Use it as an MCP server

Your agent can search 238 declassified files by location, agency, era, or type — and pull the direct war.gov source link for each.

Claude Desktop / any MCP client — add to your config:

{
  "mcpServers": {
    "uap-pulse": { "command": "npx", "args": ["-y", "uap-pulse"] }
  }
}

That's it — no API key, no auth, no cost. The data ships with the package.

Tools

Example — "What UAP files mention the Moon?" → search_sightings({ location: "Moon" }) → NASA Gemini/Apollo records with their war.gov links.

---

The live map

A 3D globe of every geolocatable record — points pulse where the sightings cluster, colored by agency, filterable by era (WWII Foo Fighters → the Navy Era). Click a hotspot to read the files and open them on war.gov. Off-world records (Moon, orbit) get their own panel.

Run it locally:

cd site && python3 -m http.server 8080   # then open http://localhost:8080

Or deploy the site/ folder to Vercel (static, no build step).

---

Data

• Source: war.gov/UFO — PURSUE Releases 1 & 2 (2026-05-08, 2026-05-22).
• License of the data: Public domain (U.S. federal government work, 17 U.S.C. § 105).
• Coverage: 238 records · 186 geolocated · 1944–2026 · agencies: War (131), FBI (57), NASA (22), State (7), CIA (1), ODNI (1), DOE (3).
• Record index mirrored by the community (vfp2/pursue-ufo-files); coordinates are representative centroids for each named location, geocoded in scripts/build_data.py.

Rebuild the dataset: npm run data (or python3 scripts/build_data.py).

---

Develop

npm install
npm run build      # tsc -> dist/
npm start          # run the MCP server over stdio

License

Code: MIT © David Mosiah (@delx369). Data: U.S. public domain.