Independent AI intelligence Two editions daily · ET
Fervor AI
Menu

Analysis · October 13, 2026 · repo

bilawalsidhu/gods-eye-viewmcpagent-infrastructure

God's Eye View Is MIT Licensed. Its Data Isn't Yours to Sell.

A trending open-source globe of live planes, ships and satellites, with a voice agent and a plug-in for Claude and Codex, shows why the license badge is the smallest license in a data-heavy repo.

The most useful file in a trending open-source repo is often the one nobody opens.

People see "MIT" on a GitHub page and read it as permission for everything: fork it, ship it, charge for it. For code, that's roughly right. But a growing share of the AI projects on the trending boards aren't mostly code. They're code wrapped around data: map tiles, flight feeds, datasets, scraped archives. And the MIT license on the code says nothing about any of that.

God's Eye View is a clean example, and an unusually honest one. It's a 3D globe in your browser that pulls in thousands of live aircraft, ships, satellites and public cameras, lets you steer it by voice through an AI agent, and doubles as an MCP server so Claude or Codex can use it as a tool. It was #9 on Trendshift's daily board when I read it around 3 PM ET on October 13 (the board moves through the day), and GitHub shows about 55,000 stars.

Its code is MIT. Its map is a patchwork of licenses, some of which forbid commercial use outright. The author documented every one of them in a file called DATA_SOURCES.md. Read that file before you build anything on this repo, and learn the habit for every repo like it.

What God's Eye View actually is

The README pitches it as a spy-satellite simulator in your browser, until you notice the sources are public and the data is real (README).

That's accurate. Out of the box it overlays:

  • 11,000+ live aircraft with route history, from OpenSky and adsb.lol
  • thousands of ships worldwide, through AISStream (live ships need a free key)
  • an 838-object satellite catalog from CelesTrak
  • the last 24 hours of earthquakes from the USGS
  • about 3,900 public traffic and city cameras, projected into the 3D scene

One honest detail up front: the road traffic layer is simulated. The README says those are simulated vehicles on OpenStreetMap roads and that "individual vehicle positions are not live observations."

The AI layer has two parts. Voice control runs on the OpenAI Realtime API (OpenAI's low-latency speech interface), and the README describes "thirty tools, four jobs": moving the camera, annotating, answering analyst questions and operating the console. And the whole app can run as an MCP server, a small program that exposes the globe's tools to an AI assistant through the Model Context Protocol, the standard Claude Desktop, Claude Code and Codex use to plug in outside tools (MCP setup guide).

The latest release is v0.2.1, from October 3.

The license badge covers less than you think

Here's the sentence that matters, from DATA_SOURCES.md: "the MIT grant covers the source code only" (DATA_SOURCES.md).

Everything the globe actually shows you keeps its own terms. The file sorts them into groups, and three groups matter for anyone thinking about a product.

Data bundled inside the repo with a non-commercial license. The TeleGeography submarine cable map ships in the repo under CC BY-NC-SA 3.0, a Creative Commons license whose "NC" means non-commercial. The file's guidance is direct: "If you use God's Eye View commercially, delete src/data/local_data/telegeography_submarine_cables/." A Nepal flood scenario (the Bhote Koshi event pack) is CC BY-NC 4.0, and commercial users have to get separate permission or remove both the data and the scene.

Data bundled with share-alike terms. The datacenter, dam and military-area name layers are extracts from OpenStreetMap under ODbL 1.0, the Open Database License. Commercial use is allowed, but share-alike applies to the data: if you build and publish a derived database from it, that database has to carry the same license. Think of it like a recipe card that says "you can sell the cake, but if you publish a new version of this recipe, it has to be free to copy too."

Live feeds with their own rules. These are fetched at runtime, so they aren't in the repo, but they are in your product the moment you deploy it. The OpenSky Network flight data is under a "non-commercial research/education license," and the file notes operational use may need "a prior written agreement with OpenSky." Google News RSS is limited to "personal, noncommercial use." The adsbdb route data "may not be copied, published, or incorporated into other databases without the explicit permission." Google Maps content, used for the photorealistic 3D tiles, "may not be cached, stored, rehosted, or committed."

Plenty of the list is wide open. The USGS earthquake data, NASA imagery and NOAA weather are U.S. public domain. Several camera feeds are CC BY 4.0 with attribution required. But the planes, the headline feature, come partly from a source that says non-commercial.

Why this is the pattern to watch

Picture the path most people take with a repo like this. It's trending, it's MIT, it looks incredible in a demo. Someone at a logistics startup clones it on a Friday, swaps in their branding, and puts it in front of customers as a "live operations view" by Wednesday. Every line of code they shipped was legally fine. The cable layer, the flood scene and the flight feed may not be.

That isn't a criticism of this repo. God's Eye View is the good case: the author wrote the terms down, marked the non-commercial items, and told you which folder to delete. Most data-heavy repos don't. They ship a CSV someone scraped, or a model trained on a dataset with a research-only license, under a cheerful MIT badge, and nothing in the repo tells you otherwise.

AI makes this more common, not less. Agent tools and MCP servers are thin code wrapped around a data source or an API. The code is the easy part to license. The data is where the terms live.

Put this into practice

Run it yourself first. You'll need Node.js 24.14 or later (or 26). The README's steps:

git clone https://github.com/bilawalsidhu/gods-eye-view.git
cd gods-eye-view
npm ci
npm run doctor
npm run dev

Then open http://localhost:4173. The doctor script checks your setup and warns if you're on an end-of-life Node version. You can start with no accounts or keys at all. Add a free token from Cesium ion (the 3D map tile service) for photorealistic 3D, an AISStream key for live ships, and an OpenAI API key for voice control. Note that the Cesium ion free token is for "eligible personal, non-commercial use," which is one more term to track if this ever leaves your laptop.

Connect it to Claude Code as a tool. With the dev server running (the MCP server reads from it), the setup guide gives this command, using the absolute path to your checkout:

claude mcp add gods-eye-view -- node /abs/path/gods-eye-view/server/mcp/stdio.js --api-base http://localhost:4173

The guide also has a JSON block for Claude Desktop and a config entry for Codex. Restart the client fully and start a new chat. If you want the globe to render inside the chat, run npm run build:panel first (MCP setup guide).

Before anything commercial, run a ten-minute data check. The same check works on any data-heavy repo:

  1. Look for a data or third-party notices file: DATA_SOURCES.md, NOTICE, THIRD_PARTY, ATTRIBUTION. If there isn't one, treat every dataset as unlicensed until you prove otherwise.
  2. List what's bundled. Folders like data/, public/, assets/ and fixtures/ often hold data with different terms than the code.
  3. List what's fetched live. Search the code for the hostnames it calls; each one is a terms-of-service you're agreeing to on your users' behalf.
  4. Mark each source as open, attribution-required, share-alike, non-commercial or proprietary.
  5. Remove or replace anything non-commercial before launch, and add the attributions the rest require.

For God's Eye View specifically, the file does most of this for you: delete the TeleGeography folder, drop the Bhote Koshi scene, replace or license the OpenSky feed and Google News RSS, and keep OSM attribution on the ODbL layers.

Honest limitations

This piece is based on reading the repo's files, not on running it under load, and the repo doesn't claim to be production software. The only performance number in the README is a single 1.86-second median cold start on an M5 Mac in Chrome, which the author calls "a comparison baseline, not a guarantee for your machine or connection."

"Start without an account or API keys" is true, but the best-looking parts aren't keyless. The photorealistic globe needs a Cesium ion or Google key, live ships need AISStream, and voice needs OpenAI. The Google route is metered, and Street View is billed per panorama load.

There's an experimental option to use a ChatGPT or Codex login instead of an OpenAI API key for voice. The README says plainly that "a supported third-party Realtime contract for reusing Codex tokens has not been confirmed." In plain terms, nobody has confirmed that reusing a Codex login this way is supported. I'd stay on the API key path.

The MCP setup guide itself skips security, but the README covers it, and it's worth reading before you share anything. The server binds to localhost by default. Sharing on your network takes an explicit --host 0.0.0.0 flag, and the README warns that "a LAN-visible server brokers your configured API keys to anyone who can reach it," which means anyone on your network can run requests on your keys and your bill. For remote access it says to put "a separately reviewed authentication proxy" in front, and the full threat model lives in SECURITY.md.

The README also says the data "may be delayed, incomplete, modeled, inferred, or wrong" and rules out uses like flight or maritime navigation and emergency response. Take that literally. It's a map for looking, not for steering.

And my reading of the licenses is not legal advice. DATA_SOURCES.md ends with the right instruction: check each provider's current terms before relying on them. Terms change, and the file is a snapshot.

The badge is the beginning

God's Eye View is worth an afternoon. It's a striking demo of what an agent with good tools can do on a live map, and the MCP server makes it a fun thing to point Claude at.

It's also the best teaching example on the trending board this week, because it tells you the truth in a file most people will never open. Open it. Then go open the equivalent file in the last data-heavy repo you shipped, if there is one.

Sources: bilawalsidhu/gods-eye-view README · DATA_SOURCES.md · docs/MCP_SETUP.md · LICENSE · Releases (v0.2.1, October 3, 2026) · Trendshift daily board, read about 15:15 ET on October 13, 2026. Repo files reviewed on the main branch on October 13, 2026.

Medium metadata

Title: God's Eye View Is MIT Licensed. Its Data Isn't Yours to Sell.

Alternates to test:

  • An MIT License Doesn't Make Your Data Commercial
  • The MIT Badge on This Viral AI Globe Doesn't Cover the Planes
  • Your Next Open Source AI Repo Ships Data, Too. Check Its License.

Subtitle options:

  1. A trending open-source globe of live planes, ships and satellites shows why the license badge is the smallest license in the repo. (picked)
  2. The code is MIT. The flight feed is non-commercial. Here's the ten-minute check before you ship.
  3. What one honest DATA_SOURCES.md file teaches about building on data-heavy AI repos.

Description: God's Eye View's code is MIT, but its flight, cable and map data carry their own terms. How to run it, and what to check before shipping.

Tags (priority order):

  1. Open Source: the license lesson applies to every repo.
  2. AI Agents: the voice agent and MCP server are the draw.
  3. MCP: readers searching for MCP servers to try.
  4. Data: the data-license angle is the insight.
  5. Software Development: broad builder reach.

Article type: A (spike): a real trending repo with a clear payoff, plus a reusable check.

Publish plan: Post at 7 to 8 AM ET, let Medium distribute for 2 to 6 hours, send to subscribers only if it is performing.