For AI agents
Wire socialAF Character Studio into Claude Desktop, Cursor, OpenClaw, Codex, or any MCP client. Every skill runs on the user’s persistent character, not a stock avatar library. That’s the wedge.
Install
Get an API key from Settings, then drop the config below into your MCP client. Same shape for Claude Desktop, Cursor, OpenClaw, and any other stdio-MCP client.
~/Library/Application Support/Claude/claude_desktop_config.json
{
"mcpServers": {
"socialaf": {
"command": "npx",
"args": ["-y", "@socialaf/mcp-server"],
"env": { "SOCIALAF_API_KEY": "saf_live_..." }
}
}
}Same config shape. Drop into ~/.cursor/mcp.jsonor your client’s equivalent.
{
"mcpServers": {
"socialaf": {
"command": "npx",
"args": ["-y", "@socialaf/mcp-server"],
"env": { "SOCIALAF_API_KEY": "saf_live_..." }
}
}
}Any client that speaks stdio-transport JSON-RPC works. Spawn npx -y @socialaf/mcp-server, pass SOCIALAF_API_KEY in env.
What agents produce
Stills, video, voice, talking avatars. Same library the dashboard uses. The agent picks the format; the character stays locked.
01
QWEN 2 Pro · Flux 2 · Nano Banana 2
02
Kling v3 · Veo 3.1 · Wan 2.7 · Happy Horse
Hey, it's Alemap. I'm walking through Tokyo at golden hour, and the light is doing the thing right now…
03
Text-to-Speech · MiniMax
04
P Video Avatar · Talking Head
What agents can do
Same models, same credits, same characters as the dashboard. The agent talks to the same v1 API your code does.
Stock avatars don't build a brand. These skills run on the user's persistent character, not a rented face.
apply_marketing_preset
Pick one of 8 ad formats (Product Review, Unboxing, Talking Head Ad, Tutorial, Virtual Try-On, Hero Shot, UGC Selfie, TV Spot). Bakes in the right model, prompt template, aspect ratio, and duration for the user's character + product.
list_marketing_presets
List the 8 presets and what each one does.
30+ underlying models routed via the PixelDojo platform. Image is synchronous; video, lip-sync, and voice are async with a built-in poll-and-wait skill.
generate_image
Synchronous image (10-30s).
generate_video
Async video, returns a jobId.
generate_lipsync
Async talking-head avatar.
generate_voice
Async TTS voiceover.
wait_for_generation
Poll a job until it settles. Use after async generations.
get_generation
Poll one job by id.
list_generations
List recent jobs.
Manage characters and other asset kinds (locations, objects, outfits, styles). Reference images run through synchronous moderation on attach.
list_characters
List the user's library.
get_character
One character with reference images.
create_character
New character (or location, object, outfit, style).
attest_character
Affirm the character is an adult. Required before generating.
create_upload_url
Presigned R2 upload URL for a reference image.
attach_character_reference
Attach an uploaded image and read its moderation verdict.
remove_character_reference
Soft-delete a reference image.
delete_character
Soft-delete an asset.
Look up what's available and what it costs before submitting work.
list_tools
Image + video model catalog with aspect/duration/ref capacity.
get_billing_status
Plan, credit balance, and subscription period.
Example flow
User prompt
“Make a 9:16 product review for my Luna character holding the matte black water bottle.”
list_characters
Find Luna’s id.
apply_marketing_preset
{ presetId: 'product-review', characterId, productDescription, productImageUrl } → returns a video jobId.
wait_for_generation
Poll until status is completed.
(deliver)
Return resultUrls[0] to the user.
The wedge, made visible
Higgsfield, Arcads, HeyGen ship stock avatar libraries. We don’t. Pick a character, and every skill returns the same face.
In motion
Pick a character. Watch them move. Same face every clip, every model, every aspect ratio.
Now playing·Mia·Kling Pro 3