data.md

Use the data

The catalog is published as machine-readable files on every deploy. The metadata is CC0: copy it, bundle it or build on it without asking. The assets it links to keep their own licences.

The licence facts are the catalog's reading of each source on the entry's verified date, not legal advice and not a guarantee that the terms are still the same. Re-check the source before you ship on one.

Files

File	What it is
data.json	The whole catalog: every entry with its licence fields, plus categories, guides, stacks and counts
data.schema.json	JSON Schema (draft 2020-12) for data.json. The build fails if data.json does not match it
plugin-feed.json	A slim catalog for editor plugins, with engine fit already worked out. See Plugin feed
plugin-feed.schema.json	JSON Schema for plugin-feed.json. The build fails if the feed does not match it
catalog.csv	One row per entry, for spreadsheets. Lists are joined with ;
feed.xml	Atom feed of newly added sources
reverified.xml	Atom feed of entries whose licence was most recently re-read
category/<id>/feed.xml	Atom feed of new sources in one category, such as category/audio/feed.xml; each category page links its own
llms.txt	A plain-text summary for language-model tools

These URLs always serve the latest build. To pin a version, download the same files from a tagged release. Releases are tagged vYYYY.MM.DD, and the attached data.json carries the same string as version.

Fields

Each entry in data.json has these fields. catalog.csv has the ones marked CSV. Its column order is fixed and new columns are only ever appended. A CSV cell that starts with =, +, -, @, a tab or a carriage return gets a leading ', so a spreadsheet does not run it as a formula; strip it if you compare values.

Field	CSV	Meaning
id	yes	Short kebab-case id, unique across the catalog
name	yes	Display name
url	yes	The source's own page
category	yes	One of the ten category folders, such as 3d, audio or tools
subcategories	yes	Finer grouping within the category
license	yes	The licence as the catalog reads it. A value from the vocabulary, explained in licenses.md
license_spdx	yes	SPDX identifier, where one fits
commercial	yes	true, false, varies (per file) or unknown
attribution_required	yes	true, false or unknown
attribution_string	yes	The credit line as the source words it, where it gives one
publisher	yes	Who publishes the source, where recorded
formats, tags	yes	File formats and search tags
added	yes	Date the entry joined the catalog
verified	yes	Date the licence was last read at the source. A read date, not a validity date
status	yes	active, needs-review (an open question, read the Notes) or deprecated
page	yes	The entry's page on the site
summary		The first paragraph of the entry
path		The entry's file in this repository
thumb		{ path, width, height, credit? } of the preview image, or null
licenseFamily		Coarse grouping used by the site's filter: public-domain, cc-by, permissive, copyleft, custom and so on. Labels are in filters.licenseFamilies
attributionClass, licenseRank		Derived sort keys used by the site
maintenance, engine_lock, ue_versions, limits, grid_dimensions, camera_perspective, hardware_tags		Present only where they apply; see the site README

Top-level keys are generatedAt, version, site, categories, guides, stacks (id and title of each starter stack), featured, stats (counts of total, active, needs-review, deprecated, listed (total less deprecated) and commercial entries), filters (licenseFamilies and formatGroups, the labels and groups behind the site's filters) and entries.

Field names are stable, and data.schema.json states them as a JSON Schema the build enforces, so a field cannot be renamed or retyped without the schema changing too. New optional fields may appear, so ignore keys you do not know rather than failing on them (the schema itself rejects unknown keys, which is how the build catches drift).

Plugin feed

plugin-feed.json is for editor plugins, starting with the Godot plugin. It lists the active and needs-review entries with the fields a plugin shows, filters and credits by. That makes it about a fifth of the size of data.json once gzipped.

• format is a whole number that goes up only on a breaking change: a field removed, renamed or retyped. New fields can appear in any build, so ignore keys you do not know. A plugin that sees a higher format than it supports should keep its cached copy and ask the user to update.
• version is the same vYYYY.MM.DD string as data.json.
• Optional fields (license_spdx, publisher, credit_line, verified) are always present, as null when unset.
• works_in lists the engines (godot, unity, unreal) an entry can be used in, based on its licence (engine_lock) and formats. made_for lists the engines it is built for (tagged, native format or locked to it); the site sorts these first when you pick an engine.
• credit_line is the recorded credit line, with {…} fields the user fills in. With attribution_required and attributionClass (where notice means carry the licence text with the files), it is enough to write the same CREDITS file the site's shortlist exports.

Examples

All CC0 audio that needs no credit, with jq:

curl -s https://tmhsdigital.github.io/Free-Game-Dev-Assets/data.json \
  | jq -r '.entries[] | select(.category=="audio" and .license=="CC0" and .attribution_required==false and .status=="active") | "\(.name)\t\(.url)"'

Everything usable commercially, as a table, with pandas:

import pandas as pd
df = pd.read_csv("https://tmhsdigital.github.io/Free-Game-Dev-Assets/catalog.csv")
ok = df[(df.status == "active") & (df.commercial.astype(str).str.lower() == "true")]
print(ok[["name", "license", "attribution_required", "attribution_string", "url"]])

Follow new sources in any feed reader by subscribing to https://tmhsdigital.github.io/Free-Game-Dev-Assets/feed.xml.

From a shell, CI or an assistant

site/cli/fgda.mjs queries the same data with the site's search and filters, no install:

npx github:TMHSDigital/Free-Game-Dev-Assets search "pixel font" --license "SIL OFL" --commercial
npx github:TMHSDigital/Free-Game-Dev-Assets check kenney --engine godot   # exit 1 if it does not fit
npx github:TMHSDigital/Free-Game-Dev-Assets get incompetech --json

check exits 1 when the entry is not for commercial use, is locked to another engine, varies per file, or is still needs-review (with its open question), so a build can stop on it. In a clone, node site/cli/fgda.mjs does the same and reads site/dist/data.json when a build is there; --data <file or https URL> picks another copy.

site/cli/mcp.mjs is a Model Context Protocol server (stdio, no dependencies) with the tools search_assets, get_asset, check_license and list_stacks. In Claude Desktop, Cursor or another MCP client, from a clone:

{ "mcpServers": { "free-game-dev-assets": { "command": "node", "args": ["/path/to/Free-Game-Dev-Assets/site/cli/mcp.mjs"] } } }

How to cite

The repository has a CITATION.cff, and GitHub's "Cite this repository" button renders it. In text:

TMHSDigital. Free Game Dev Assets: a dated, sourced catalog of free game assets and tools with licence metadata (version vYYYY.MM.DD) [Data set]. https://github.com/TMHSDigital/Free-Game-Dev-Assets

@misc{free_game_dev_assets,
  author       = {TMHSDigital},
  title        = {Free Game Dev Assets: a dated, sourced catalog of free game assets and tools with licence metadata},
  howpublished = {\url{https://github.com/TMHSDigital/Free-Game-Dev-Assets}},
  note         = {Dataset, version vYYYY.MM.DD},
  license      = {CC0-1.0}
}

Replace vYYYY.MM.DD with the release you used.