bitHuman docs

npx add-mcp 'https://docs.bithuman.ai/docs-mcp'

For AI agents

URL: https://docs.bithuman.ai/resources/agents

Everything an AI agent needs to read these docs and build with bitHuman.

Everything on this site is available in a form an agent can read without parsing HTML.

Build with a coding agent

Paste this into Claude Code, Cursor, Codex or another coding agent, with your project open:

Read https://docs.bithuman.ai/llms.txt and add a bitHuman avatar to this project: <what you are
building, for example "an iPhone app where a cartoon fox answers questions">. Pick essence-2 for a
real person or expression-2 for any other character, start from the matching example in
https://gitlab.com/bithuman/sdk/bithuman-examples (its llms.txt lists them), and read my API secret
from BITHUMAN_API_SECRET. Never print it, put it on a command line or commit it.

You need an API secret on the Creator plan or higher: sign up, then choose a plan. For better results, also give the agent the docs MCP server and the agent skill below.

Markdown for every page

• Add .md to any page's URL: /platforms/python.md. Each HTML page links its twin with rel="alternate".
• Or ask for markdown: a request with Accept: text/markdown gets the same text at the page's own URL.

curl -H "Accept: text/markdown" https://docs.bithuman.ai/platforms/python

Files

File	What it holds
/llms.txt	The index: rules for agents, where each model runs, key facts, one command per platform, speed, and links
/llms-full.txt	Getting started, the Python, CLI, Windows, LiveKit, Pipecat and REST platform pages, and the REST API in one file
/llms/start.txt, /llms/platforms.txt, /llms/apps.txt (iOS & iPadOS, macOS, Android, Flutter, Web), /llms/deploy.txt, /llms/models.txt, /llms/build.txt, /llms/api.txt	The same text one section at a time
/api/openapi.yaml	The REST API as OpenAPI 3
/versions.json	The current version and install line of every artifact
/platforms.json	Every way to run bitHuman, with its first command and docs
/performance.json	The speed of every published configuration (× real time), with hardware, release and date
/changelog.xml	Every release as an RSS feed, tagged by platform
Showcase	Sample avatars anyone can use: slug, agent code, model

Docs MCP

The docs are also an MCP server with two read-only tools: search finds pages, and fetch returns one as markdown. It needs no account and no key.

Client	Add it
Claude (claude.ai, desktop)	add bitHuman docs from the Connectors Directory
Claude Code	claude mcp add --transport http bithuman-docs https://docs.bithuman.ai/docs-mcp
Cursor	Add to Cursor (opens Cursor), or add the JSON below to ~/.cursor/mcp.json
VS Code	Add to VS Code (opens VS Code), or code --add-mcp '{"name":"bithuman-docs","type":"http","url":"https://docs.bithuman.ai/docs-mcp"}'
Other clients	a remote (Streamable HTTP) server at https://docs.bithuman.ai/docs-mcp

{ "mcpServers": { "bithuman-docs": { "url": "https://docs.bithuman.ai/docs-mcp" } } }

It is also on the official MCP Registry as ai.bithuman/docs.

To drive bitHuman itself (create agents, render videos) from an MCP client, use the CLI's own server instead: MCP server. To have bitHuman's own characters speak inside ChatGPT or Claude, add the hosted server https://mcp.bithuman.ai/mcp (how).

Agent skill

bithuman-integrate is a skill an agent loads before it adds an avatar to a project: it picks the path, installs, wires 16 kHz speech to frames and handles the API secret. Save it as .claude/skills/bithuman-integrate/SKILL.md in a project, or ~/.claude/skills/bithuman-integrate/SKILL.md for every project:

mkdir -p .claude/skills/bithuman-integrate
curl -fsSL https://docs.bithuman.ai/skills/bithuman-integrate/SKILL.md -o .claude/skills/bithuman-integrate/SKILL.md

Rules for agents

1. Send model ("essence-2" or "expression-2") on every agent you create, and poll until status is ready or failed.
2. API and SDK use requires the Creator plan or higher.
3. Read the API secret from the environment (BITHUMAN_API_SECRET); never write it into code or a command line. In a LiveKit worker the secret is read as BITHUMAN_MASTER_SECRET (the same secret under another name, so the plugin never copies it into the room) and the plugin gets a minted token (LiveKit).
4. When explaining, say where the avatar renders and where the conversation runs: the avatar renders (device, browser, server, cloud); the conversation runs (your stack, the CLI's local conversation brain, or bitHuman's servers).
5. Take versions from /versions.json, speed from /performance.json and prices from GET https://api.bithuman.ai/v1/pricing (send the api-secret header; without it the call returns 401), not from memory or an older sample.

Next

• MCP server (the CLI) · Quickstart · API reference