AvatarSay MCP Server
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| AVATARSAY_API_KEY | No | Optional AvatarSay API key (avs_...) from https://avatarsay.com/dashboard. Public tools (list_voices_and_styles, roll_rpg_npc) run with 0 credits and no API key; pass an API key to generate portraits, MP3 voice lines, and talking MP4 videos. |
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
AvatarSay generates original fictional character portraits (2 × 1024×1024) in 8 art styles, imports verified stylized character illustrations (anime, cartoon, pixel art, fantasy/VTT tokens), rolls deterministic tabletop RPG NPCs with SillyTavern V2 & Foundry VTT JSON exports, renders fast (<3s) MP3 character voice barks, and animates lip-synced square talking videos (960×960 MP4, 5–30s, 8 English voices, no watermark). Free & TRPG workflow: (1) call roll_rpg_npc (public, 0 credits) to roll a complete D&D / tabletop NPC with SillyTavern Card V2 + Foundry VTT Actor JSON and a <=75-char greeting; (2) call generate_avatars (4–6 credits) and poll get_job after ~5–10s for 2 portrait assetIds, OR call import_character_portrait (0 credits, confirmFictional=true) to import an existing stylized character illustration; (3) call create_npc_voice_line (2 credits/100 chars) for an instant MP3 voice bark, or call create_talking_video (first <=5s / <=75-char clip is 0 credits on eligible new accounts!) and poll get_job every 20–30s for the MP4 download URL. Safety rules: stylized fictional characters only — camera photographs, photorealistic human likenesses, celebrities, impersonation, and explicit/harmful content are strictly blocked.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {
"listChanged": false
} |
| prompts | {
"listChanged": false
} |
| resources | {
"subscribe": false,
"listChanged": false
} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| list_voices_and_stylesA | Return all 8 AvatarSay visual styles, all 8 built-in English TTS voices (with audio sample URLs), per-action credit costs (including first free 5s talking clip and 2-credit NPC voice lines), and one-time credit pack pricing. Public read-only tool — does not require an API key. |
| roll_rpg_npcA | Deterministically roll a complete tabletop RPG / D&D NPC (name, ancestry, role, traits, quirk, plot hook, GM secret, portrait prompt, matched voice sample MP3, and a <=75-char greeting tuned for the free 5s talking video clip), plus ready-to-import SillyTavern Character Card V2 JSON and Foundry VTT NPC Actor JSON. Public zero-cost tool — requires NO API key and charges 0 credits. |
| get_creditsA | Return the authenticated AvatarSay account email, current non-expiring credit balance, image quality tier (standard vs hd), whether the 0-credit first talking video clip is still available, and how many avatar generations, NPC voice lines, or video seconds the balance covers. Requires Authorization: Bearer avs_... |
| list_avatarsA | List recent fictional character avatars (both AI-generated portraits and imported stylized character illustrations) in this AvatarSay account, including assetId (ast_...) and 24-hour signed download URLs, ready to animate with create_talking_video. Requires Authorization: Bearer avs_... |
| import_character_portraitA | Import an existing non-photorealistic fictional character illustration (anime, 2D/3D cartoon, pixel art, fantasy digital painting, VTT token art; PNG/JPG/WebP, 256–4096 px, max 5 MB) from an https:// URL or base64 data URI for 0 credits, returning an assetId (ast_...) ready for create_talking_video. Camera photographs and photorealistic human likenesses are strictly blocked by EXIF and automated vision screening. Requires confirmFictional=true and Authorization: Bearer avs_... |
| list_videosA | List recent watermark-free talking avatar MP4 videos (960×960) already rendered in this AvatarSay account, including assetId (ast_v_...), spoken script, and 24-hour signed stream & download URLs. Requires Authorization: Bearer avs_... |
| generate_avatarsA | Generate 2 square (1024×1024) fictional character portraits from a text description in one of 8 art styles. Costs 4 credits (Standard) or 6 credits (HD for accounts that have purchased a pack). Enqueues asynchronously and returns jobId immediately — call get_job after ~5–10 seconds to get the 2 assetIds and signed image URLs. Requires Authorization: Bearer avs_... |
| create_npc_voice_lineA | Synthesize an instant (<3s) English character voice bark or tabletop NPC dialogue MP3 (1–500 chars) in one of 8 preset voices with optional speed and emotion control. Costs 2 credits per 100 characters (minimum 2 credits — a new 30-credit account can generate 15 tabletop MP3 barks free). Returns a 24-hour signed MP3 download URL immediately. Requires Authorization: Bearer avs_... |
| create_talking_videoA | Animate an AvatarSay character avatar (ast_...) with lip-synced English speech in one of 8 preset voices. New accounts get their first short talking clip (<=5s / <=75 chars) for 0 credits! Subsequent or longer clips cost 12 account credits per second of speech (minimum 5s = 60 credits, maximum 30s). Enqueues asynchronously and returns jobId immediately — never blocks for the 3–5 minute render. Poll get_job every 20–30 seconds for the watermark-free 960×960 MP4 URL. Failed renders refund credits automatically. Requires Authorization: Bearer avs_... |
| get_jobA | Check the status (queued, running, succeeded, failed) of an image or talking-video job (job_...) and return 24-hour signed download URLs once succeeded, or refund details if failed. Requires Authorization: Bearer avs_... |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
| create-rpg-npc | Roll a complete tabletop RPG NPC with SillyTavern V2 & Foundry VTT JSON exports, generate 1024×1024 portraits, and render an MP3 voice bark or free 5s talking MP4. |
| prep-tavern-encounter | Roll 3 interconnected tabletop NPCs with Foundry VTT Actor JSON & SillyTavern cards and synthesize instant MP3 voice lines for session-zero prep. |
| create-talking-intro | Generate an original fictional character avatar and animate a short spoken intro clip. |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
| AvatarSay Styles, Voices & Credit Pricing | Live catalog of 8 art styles, 8 English voices, TRPG tools, credit costs, and packs (https://avatarsay.com/mcp). |
TDQS
Scored across 10 tools
Each tool maps to a distinct resource+action: public catalog reads, avatar generation vs. portrait import, NPC voice synthesis, talking video rendering, job polling, and credit/asset listing. The only mild overlap is that both list_voices_and_styles and get_credits surface credit/pricing information, and both generate_avatars and import_character_portrait produce avatars (though via clearly different paths). Descriptions make the boundaries clear enough that an agent should select correctly.
All ten tools follow a consistent snake_case verb_noun pattern: list_voices_and_styles, roll_rpg_npc, get_credits, list_avatars, import_character_portrait, list_videos, generate_avatars, create_npc_voice_line, create_talking_video, get_job. Verbs (list/get/create/generate/import/roll) map predictably to read vs. mutate semantics.
Ten tools is well within the ideal range and each one earns its place: two public onboarding tools, three read tools, three creation/import tools, and a shared job poller. No filler or redundant variants are present.
The core lifecycle is covered end-to-end: discover voices/styles and costs, acquire avatars (generate or import), synthesize voice lines, render talking videos asynchronously, and poll job status with refund handling. Gaps are minor but real — no delete/cleanup for avatars or videos, no job cancellation despite 3–5 minute renders, and no credit purchase tool despite pricing being advertised.