Remote Model Context Protocol · Streamable HTTP

AI Talking Character MCP Server

Point Claude, Cursor or your own agent at one URL to design original D&D NPCs, indie game portraits and anime characters — then turn any portrait into a lip-synced square MP4 video with a built-in voice.

Live endpoint https://avatarsay.com/mcp

Get API key in Dashboard See one-time credit packs

No local Python or Docker install. New accounts start with 30 free credits (7 avatar generations = 14 portraits). One-time packs start at $9.90 and never expire.

Setup in 60 seconds

Connect Claude, Cursor or Windsurf

Sign in once at /dashboard, open the MCP & API keys tab, and create a personal avs_… key. You can also connect without a key first to inspect tools and browse voices and styles.

Cursor / Windsurf

Paste into .cursor/mcp.json or Windsurf MCP config:

{
  "mcpServers": {
    "avatarsay": {
      "url": "https://avatarsay.com/mcp",
      "headers": {
        "Authorization": "Bearer avs_YOUR_API_KEY"
      }
    }
  }
}

Claude Code (CLI)

Add the remote Streamable HTTP server in one terminal command:

claude mcp add --transport http avatarsay https://avatarsay.com/mcp --header "Authorization: Bearer avs_YOUR_API_KEY"

Claude Desktop (stdio bridge)

For desktop clients that use claude_desktop_config.json:

{
  "mcpServers": {
    "avatarsay": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://avatarsay.com/mcp",
        "--header",
        "Authorization:${AUTH_HEADER}"
      ],
      "env": {
        "AUTH_HEADER": "Bearer avs_YOUR_API_KEY"
      }
    }
  }
}

Tools & async design

Six focused MCP tools, zero blocking timeouts

Talking avatar video renders take 3–5 minutes (9–14 minutes for 15–30 s clips). Instead of holding an HTTP connection open until the client times out, generate_avatars and create_talking_video enqueue the job immediately and return a jobId for your assistant to poll with get_job.

Tool name	Auth	Credits	What it does
list_voices_and_styles	Public (no key)	Free	Returns all 8 visual styles, all 8 English TTS voices (with audio sample URLs), credit costs, and pack pricing.
get_credits	Bearer avs_…	Free	Checks your current non-expiring credit balance, image tier (Standard vs HD), and remaining avatar sets or video seconds.
list_avatars	Bearer avs_…	Free	Lists recent fictional character avatars in your account with assetId (ast_…) and 24-hour signed image URLs.
generate_avatars	Bearer avs_…	4 Standard / 6 HD credits	Generates 2 × 1024 px portraits from a text prompt and style. Enqueues asynchronously and returns jobId immediately.
create_talking_video	Bearer avs_…	12 credits/s (min 60)	Synthesizes English speech and queues a lip-synced 960×960 MP4 (5–30 s, no watermark). Returns jobId immediately — never blocks for the 3–5 min render.
get_job	Bearer avs_…	Free	Polls an image or video job (queued → running → succeeded / failed) and returns 24-hour signed download URLs or automatic refund details.

Preset styles

8 character art styles

Pass any style ID below to generate_avatars:

Anime anime
3D Cartoon 3d-cartoon
Pixel Art pixel-art
Realistic realistic
Professional professional-headshot
Fantasy fantasy
Watercolor watercolor
Cyberpunk cyberpunk

Built-in English TTS

8 preset character voices

Pass any voice ID below to create_talking_video (no voice cloning):

Friendly Friendly_Person

Warm and upbeat, an easy everyday voice.
Warm Woman Wise_Woman

Warm, measured female voice.
Lively Girl Lively_Girl

Bright, energetic, younger-sounding female voice.
Calm Woman Calm_Woman

Calm, steady female voice.
Casual Guy Casual_Guy

Relaxed, conversational male voice.
Deep Voice Deep_Voice_Man

Low, resonant male voice.
Patient Man Patient_Man

Slow, unhurried male voice.
Young Hero Young_Knight

Younger-sounding, energetic male voice.

Pay once, use anywhere

No monthly plan required for MCP

Your web studio and MCP keys share the same credit balance. New accounts receive 30 free credits on sign-up — enough to run 7 generate_avatars calls (14 character portraits) via MCP for free.

Talking videos via MCP require account credits (12 credits/s, minimum 5 s = 60 credits). The one-time free 5 s trial clip is reserved for interactive web sessions on /talking-avatar so automated scripts cannot drain the trial pool.

Starter

$9.90

500 credits · one-time

≈ 83 HD avatar pairs
≈ 41 s of talking video
Never expires

Creator

$24.90

1,500 credits · one-time

≈ 250 HD avatar pairs
≈ 125 s of talking video
Never expires

Studio

$59.00

4,000 credits · one-time

≈ 666 HD avatar pairs
≈ 333 s of talking video
Never expires

Example prompts

What builders run through AvatarSay MCP

D&D & Tabletop RPGs

Session-zero NPC portraits & quest hooks

“Read my campaign notes for Marrowfen Keep, generate fantasy portraits for the three faction leaders with AvatarSay, and render a 6-second talking MP4 of Lord Vane delivering his warning.”

Indie Game Prototyping

Pixel-art & anime dialogue cutscenes

“Generate pixel-art portraits for our shopkeeper NPC, pick the first assetId, and render her opening bark in Lively_Girl voice so we can drop the MP4 into our pitch deck.”

Creator Workflows

Faceless mascot intros from code or docs

“Summarize today’s release changelog into one punchy 10-second spoken line and render a talking video using my existing 3d-cartoon mascot avatar.”

FAQ & safety

Common questions

Questions or directory verification? Email support@avatarsay.com.

Do I need a monthly subscription to use the AvatarSay MCP server?

No. Every new account gets 30 free credits on sign-up (enough for 7 avatar generations via MCP), and paid credit packs are one-time purchases starting at $9.90 for 500 credits. Credits never expire.

Can I upload a real person’s photo or clone a voice through MCP?

No, by design. AvatarSay only animates fictional characters generated inside your AvatarSay account (assetId starting with ast_). Photo uploads, face swaps, real public figures, and voice cloning are blocked across both the web app and MCP, and every prompt and script passes automated safety screening before credits are charged.

Why do generate_avatars and create_talking_video return a jobId instead of waiting?

Lip-synced video rendering takes 3–5 minutes for a short clip (up to 9–14 minutes for 15–30 second clips). Returning jobId immediately prevents Claude, Cursor and HTTP proxies from timing out. Your AI assistant polls get_job until status is succeeded, then gives you a 24-hour signed watermark-free download link. If a render ever fails, credits are refunded automatically.

How are MCP API keys stored and revoked?

Keys start with avs_ and are shown once when created. AvatarSay stores only the SHA-256 hash and a short prefix in the database. You can hold up to 3 active keys and revoke any key immediately in Dashboard → MCP & API keys.