The full upstream README, mirrored here for reference. Install config, tool schemas, adoption signals, and an original overview live on the Chenji Affect listing page.
Turn raw text into structured affect signals — an 8-dimensional emotion vector, emotion texture labels, causal intent, 3D avatar driving parameters, empathy response strategies, and somatic-sensation emotion decoding — inside any MCP-capable client (Claude Desktop, Cursor, Qoder, etc.).
Powered by the Chenji Affect API (taichusjs.cn).
| Tool | Level | Description |
|---|---|---|
analyze_text | L1 | Text → 8-dim affect vector + emotion texture + intent + state description |
generate_avatar_params | L2 | Text → blendshape/AU/curve animation package + lighting/atmosphere package + upstream L1 |
empathy_hint | L3 | Text → deterministic empathy response strategy (approach, tone, pacing, focus, avoid-list), ~20ms, no LLM |
somatic_decode | L4 | Body-sensation text → structured emotion (primary affect, valence/arousal, texture, intent) + somatic anchors |
Clients that support streamable-http MCP can connect directly, no local package needed:
Requires Python 3.10+.
Get an API key:
https://taichusjs.cn/trial.html.Environment variables:
| Variable | Required | Description |
|---|---|---|
CHENJI_API_KEY | Yes | Your Chenji API key |
CHENJI_API_BASE | No | Defaults to https://taichusjs.cn |
generate_avatar_params requires a key tier that includes L2.empathy_hint and somatic_decode are billed as L1-family calls (trial tier applies).