AGENTS

Everything on this site, for an agent: an MCP server, a skill, llms.txt and plain JSON. Point your agent here and it can make DREW without you pasting anything.

The MCP server

https://drew.audio/mcp speaks the Model Context Protocol over Streamable HTTP. It is public and read-only: no account, no key, no state.

Connect

Claude Code:

claude mcp add --transport http drew https://drew.audio/mcp

Claude (web and desktop): Settings → Connectors → Add custom connector, and paste https://drew.audio/mcp.

Cursor, Windsurf and other clients that take a JSON config:

{
  "mcpServers": {
    "drew": {
      "url": "https://drew.audio/mcp"
    }
  }
}

Codex CLI (~/.codex/config.toml):

[mcp_servers.drew]
url = "https://drew.audio/mcp"

Clients that only speak stdio can bridge with:

npx mcp-remote https://drew.audio/mcp

Tools

drew_canon

Who DREW is, in one call: the constants that never change (the black locs, the single glacier-blue loc and its silver moon cuff, the beard, the hazel-green eyes, the pendant and the watch), the identity text to paste into every prompt (a full form for close and medium shots, a short form for full length and wider), every look's wardrobe text, his world, his voice and manner, the never list, Claudia as a guest, and the norms for releasing work with him. Call this first, before making anything.

list_films

The songs with a video out. Today that is one: the D.R.E.W film (1:49.6). For each: the logline, the numbers from the generation log (shots, still steps, Seedance clips, lip-sync passes), the film's links, the Spotify link and the synced lyric. New songs are added to the same list.

browse_images

Page through the stills, filtered by set: face (portraits and a face sheet), film (start frames and stills from the film), looks (his looks outside the film), more and refs. Returns compact entries: id, set, title, caption, size, image URLs, and the prompt when the generation log has it. Use get_image for everything about one still.

search_images

Full-text search over the stills' titles, captions and prompts. All words must match. Example queries: "garage moon", "penthouse charts", "claudia car", "watch", "rooftop sign", "face sheet".

get_image

One still by id: its set, title and caption, its size, where it was used (a film shot, a look, a reference), URLs at two sizes, and the exact prompt, model and settings when the film's generation log records them for it. If the log has nothing for that still, it says so; it never makes one up. Ask for the picture itself to look at him.

read_wiki

An article of the DREW wiki as Markdown: start (making a video with him from nothing), consistency (keeping him himself), song (song to shot list), stills (references in GPT Image 2.5), seedance (moving shots and lip sync in Seedance 2.5), cars (parked cars and doors that open right), edit (cutting to the beat, every word on screen), failures (what went wrong and the rule from each), agents (making him with an agent). Leave the topic out to get the list.

A first session might go: drew_canon → search_images("garage moon") → get_image(id) → read_wiki("seedance").

The skill

The same knowledge as files, for agents that load skills (Claude Code, the Claude apps and others that read SKILL.md): the canon and the rules in SKILL.md, and the wiki and the film's record as references.

curl -sL https://drew.audio/skill/drew-skill.zip -o drew-skill.zip
unzip -o drew-skill.zip -d ~/.claude/skills/

Or read it first: https://drew.audio/skill/SKILL.md.

llms.txt and Markdown

https://drew.audio/llms.txt indexes the site for language models. The character is one Markdown file at /drew.md, and every wiki article has a Markdown twin: add .md to its path without the trailing slash (/wiki/seedance.md).

JSON

Everything is static and cacheable.

Use

DREW is an open character; see Use him. Keep agents polite: cache what you fetch, and prefer the medium images unless you need the large ones.