mcp.film / Avatars & Performance / bitHuman MCP
bitHuman MCP
Official bitHuman CLI server: build avatar agents from a face image and render WAV-driven talking video locally.
What it does
bitHuman's MCP ships inside its own CLI: `bithuman mcp` exposes agent creation from a still (generate_agent), gesture generation, model pulls and a local render that turns a WAV into a lip-synced MP4 on your machine. A separate hosted endpoint lets any chat client make bitHuman's stock characters speak a line, no account needed. For a filmmaker-agent it is a talking-head and digital-performer tool that pairs with whatever TTS or recorded dialogue you already have, since bitHuman deliberately does not synthesize speech.
Connect
https://mcp.bithuman.ai/mcp
claude mcp add bithuman -- bithuman mcp
{
"mcpServers": {
"bithuman": {
"command": "npx",
"args": [
"-y",
"mcp-remote",
"https://mcp.bithuman.ai/mcp"
]
}
}
}
{
"mcpServers": {
"bithuman": {
"url": "https://mcp.bithuman.ai/mcp"
}
}
}
Local install: Multi-step: install the CLI with `curl -fsSL https://install.bithuman.ai | sh`, authenticate with `bithuman login` (or set BITHUMAN_API_SECRET in the client config), then run the server as `bithuman mcp`
BITHUMAN_API_SECRET — Run `bithuman login`, or use the API secret from your bitHuman account; API and SDK access is listed from the Creator plan up (https://www.bithuman.ai/pricing)
Sample tools
generate_agentget_agent_statuslist_agentsagent_speakgenerate_dynamicsrenderpullupload_fileget_credit_balancelist_charactersmake_character_videoField notes
Verified 8 Oct 2026 against bitHuman's MCP docs and the official MCP Registry, where ai.bithuman/characters was published that same morning; no tools were run. Two servers share the name and are not the same thing. The hosted remote_url takes no auth and exposes only list_characters, make_character_video (3-200 characters of text) and talk_to_character — the docs say outright it 'speaks only as bitHuman's own characters', so it is a demo, not a pipeline tool. Your own avatars need the CLI server, which is the claude_code command above. There is no text-to-speech tool: list_voices and text_to_speech were retired after CLI 2.8.7 and now return 410, so bring a WAV from elevenlabs, cartesia or a real recording and pass it to render. generate_agent is async and slow — the docs put a second-generation agent at roughly 2-2.5 hours — and always pass model (essence-2 or expression-2) or you get a first-generation agent. Every model except Expression 2 needs a real human face; stylised or non-human subjects return 422 MODEL_SUBJECT_MISMATCH. The cloud talking-video API and knowledge bases have no MCP tool yet, so use the REST API for those. Pricing: Free is $0 with 99 credits a month, Creator $20 for 2,000. bitHuman's pricing page says Free-plan API access ends on 12 October, so plan on Creator. The install line pipes a remote script into sh: read it first. Source is not public — every repo in the bithuman-product GitHub org was archived on 5 Oct 2026 and the registry metadata now points at GitLab — so repo is null. For hosted, OAuth-gated performance work without a local CLI, hedra-official is the simpler connect.
Also in Avatars & Performance
Hosted MCP for avatar videos, lipsync, and 175-language video translation.
Phoenix replicas, persona videos, lipsync, and conversational AI via the Tavus API.
Character-3 talking-avatar video, TTS, and image generation through Hedra's API.
D-ID's hosted MCP connecting agents to the D-ID API and docs for avatar video.
Official CraftStory server: talking-avatar video from a photo and audio, plus short MiniMax H3 clips with sound.
VEED's hosted MCP for talking-head video: pick a character, pick a voice, send a script.
Pairs well with
Official hosted Artlist MCP — generate images, video, music and voice-over on your Artlist credits.
Hosted OAuth MCP onto BudgetPixel's image, video, music, SFX and Topaz upscale catalogue.
Hosted screenplay-to-production MCP with free planning and quotes, then a greenlit run under a server-enforced cap.