Plugin Universe

An open database of DAW plugins. Under Construction - Read Me First

← search

Services

Everything here is one catalogue, offered five ways. Pick whichever suits what you are: a person, a program, an agent, or a database.

The facts are CC0 — public domain. Attribution to Plugin Universe is requested and not required. Contributor-written prose is CC BY-SA 4.0 and must be attributed. A plugin's own licence is its author's and is a separate matter from the catalogue's.

For a person

plugin-universe.com — search by what a plugin does, not just its name. Retrieval is semantic, so "warm analogue bus compressor" works better than a product name.

A search has an address: /search?q=warm+analogue+bus+compressor is the page you are looking at, so it can be linked and shared. /plugins is the whole catalogue, most recently added first, a page at a time.

For a program: the JSON API

No key, no account, CORS open.

GET /search?q=warm+analogue+bus+compressor&limit=10
GET /search?category=reverb&pricing=Free&source=OpenSource
GET /plugins?limit=50&order=recent
GET /facets
GET /plugin/<slug>.json
GET /category/<slug>.json
GET /health

Facets are format, category, role, vendor, source, pricing, licence, measured, accepts and produces. They filter and never re-rank. They work on /plugins as well as /search.

accepts and produces are the signal types from the plugin profile — Audio, Midi, ControlMidi and the rest, as /facets lists them. They are what makes a chain answerable rather than just a catalogue: given something that produces MIDI, /search?accepts=Midi is the set of plugins that can follow it.

/search and /plugins answer in two languages. JSON is the default — that is what these paths have always been, and a caller sending no Accept header or a wildcard one still gets exactly what it got before. Ask for text/html and you get the page instead, which is what makes one of these URLs something you can paste to a colleague. /facets is JSON only: the facet values are the dropdowns on the search form, so there is no separate page to send anyone to.

For a database: content negotiation

Every plugin IRI dereferences. Ask for what you want:

curl -H "Accept: text/turtle"        https://plugin-universe.com/plugin/<slug>
curl -H "Accept: application/ld+json" https://plugin-universe.com/plugin/<slug>

Or by suffix — .ttl, .jsonld, .json. Category IRIs resolve the same way.

IRIs are minted under http://purl.org/stuff/plugin-universe/ and redirect here, so they survive this site moving. The vocabularies they use are at /ns.

SPARQL

https://sparql.plugin-universe.com/public/query

Read-only. The default graph is the union of every named graph, so a query with no GRAPH clause sees everything. Details, limits and what is deliberately absent.

For a package manager

https://plugin-universe.com/registry/plugins/index.json

An Open Audio Stack compatible registry: every plugin the catalogue has a release for, with versions, download URLs and SHA-256 checksums. OwlPlug and StudioRack can read it as-is.

For an agent: MCP

https://mcp.plugin-universe.com/mcp

Streamable HTTP, stateless, no account. Four tools: search_plugins, get_plugin, list_categories and sparql_query. What each does.

Claude Code

claude mcp add --transport http plugin-universe https://mcp.plugin-universe.com/mcp

--scope user makes it available in every project; the default is this project only. To share it with a repository, use --scope project and it is written to .mcp.json:

{
  "mcpServers": {
    "plugin-universe": {
      "type": "http",
      "url": "https://mcp.plugin-universe.com/mcp"
    }
  }
}

Check it with claude mcp list.

Codex

codex mcp add plugin-universe --url https://mcp.plugin-universe.com/mcp

Or in ~/.codex/config.toml:

[mcp_servers.plugin-universe]
url = "https://mcp.plugin-universe.com/mcp"

No bearer_token_env_var and no auth: the endpoint is open and read-only, so there is nothing to authenticate.

Anything else

Any MCP client that speaks Streamable HTTP. The endpoint answers at https://mcp.plugin-universe.com/ as well, for clients that expect the server at the root.

Two things worth knowing before you quote it

Provenance travels with the facts. Every plugin record names the source it came from and that source's licence. If you repeat a claim from here, you can say where it came from — and should.

A measurement is not a property of a plugin. It is a reading from one binary, on one machine, on one day, so the tool, the platform and the date come back with every number. "This plugin uses 2% CPU" is a claim this catalogue cannot support. "A scan on Linux x64 in September reported five ports" is one it can. What each verdict means — including which of them are about the plugin and which are about our container.

Bulk

/dumps/ — the whole dataset as files, browsable. Served straight off disk rather than through the application, so taking it costs this server almost nothing.

  • cc0/ — the factual catalogue, public domain
  • notice/ — material from permissively-licensed sources, with their notices
  • prose/ — contributor-written text, CC BY-SA, attribution required
  • MANIFEST.json names each file's graph, licence and source
  • void.ttl describes the dataset

The three parts are separate directories because they are separate datasets under separate terms; which licence applies to a file is a property of where it sits rather than something you have to work out. Graphs whose licence does not permit redistribution are not in any of them — they were never written, rather than filtered on the way out.

If you want everything, take this rather than paginating SPARQL. It is faster for you and cheaper for us, and it is the same data.

Rebuilt nightly. MANIFEST.json carries the timestamp, so a consumer that only wants a changed dataset can check that rather than fetching the files.

Please do not paginate the SPARQL endpoint to scrape the catalogue: it costs this server far more than it costs you, and there is a better answer available for the asking.

Limits, and contact

The API and the SPARQL endpoint are rate-limited, and SPARQL queries time out at thirty seconds. This runs on one small machine.

If something here is broken, insufficient, or you need access in a shape that is not offered, get in touch. A bulk consumer paginating the SPARQL endpoint costs this server far more than answering the email would.