Claude writes and runs Cavalry scripts live — no copy-pasting, no window switching.
- Write & run scripts — Claude executes Cavalry JavaScript live via the Stallion bridge
- Knows the docs — searches 40,000+ chunks of Cavalry documentation, real scenes, and community knowledge semantically
- Scene-aware — reads and saves your active scene
- Stays current — auto-loads a verified best-practices reference on every session
| Tool | Purpose |
|---|---|
| Cavalry | The app — needs Stallion enabled |
| Node.js 18+ | Runs the MCP server |
| Ollama | Local embeddings |
| Claude Code | The AI client |
git clone <repo-url> cavalry-assistant
cd cavalry-assistant
cp .env.example .env
npm install --prefix mcp
npm run build --prefix mcpDownload lancedb_public.zip from GitHub Releases, extract into the project:
cavalry-assistant/
data/
lancedb_public/ ← extract here
Update .env:
LANCEDB_PATH=./data/lancedb_publicWindows / network drive users: LanceDB requires a local drive. Set an absolute local path:
LANCEDB_PATH=C:/Users/yourname/.cavalry-assistant/lancedb_public
ollama pull nomic-embed-textScripts > Stallion — leave it running on port 8080.
The .mcp.json in this folder auto-registers when opened in Claude Code. Or manually:
claude mcp add cavalry-assistant node mcp/dist/index.js/cavalry
| Source | Content |
|---|---|
| Official docs | Full Cavalry documentation |
| Scenery scenes | 270 real .cv scene files — real-world node/connection patterns |
| Manual reference | Curated scripting patterns and best practices |
| Scripts | Example JavaScript scripts |
The public knowledge base does not include Discord community data (private).
cd etl
pip install -r requirements.txtLANCEDB_PATH=C:/Users/yourname/.cavalry-assistant/lancedb
DISCORD_TOKEN=your_token
DISCORD_CHANNEL_IDS=channel_id_1,channel_id_2
SCENERY_COOKIE=your_session_cookiecd .. # project root
# Public sources
python etl/ingest.py --source docs
python etl/ingest.py --source scenery
python etl/ingest.py --source manual
python etl/ingest.py --source scripts
python etl/ingest.py --source cv
# Private sources (stays local, never exported)
python etl/ingest.py --source discordpython etl/export_public_db.py --zip
gh release create v1.x --title "Knowledge Base v1.x" data/lancedb_public.zippython etl/scrape_scenery.py
python etl/ingest.py --source cv
python etl/export_public_db.py --zip- Stallion always responds
"Success"— check Cavalry's console panel for actual errors - Always start scene-modifying scripts with
api.stop() - Full API reference:
prompts/cavalry-best-practices.md - The public DB is a clean snapshot — no Discord messages, no credentials, no user data
Built with Cavalry by Scene Group, Stallion, Model Context Protocol SDK by Anthropic, LanceDB, and Ollama + nomic-embed-text.
Knowledge base sourced from Cavalry official docs, Scenery, and the Cavalry community.