Moose Basics

A MooseStack project is a declarative TypeScript/Python project that gets translated into a running analytics stack: HTTP ingest → Redpanda → ClickHouse → aggregation views → HTTP egress APIs. moose dev runs that stack locally and hot-reloads when you edit app/.

The sections below are independent — jump to whichever is relevant.

Golden rules — apply always

1. Use moose ls (or the mcp__moose-dev__get_infra_map MCP tool) as the source of truth for URLs and table names. Never curl or query a guessed path. After every edit to app/index.ts, run moose ls. If your route/table isn't listed, the TS compile failed — check logs (rule 3), don't keep poking.
2. Use the explicit OlapTable + Stream + IngestApi trio. Do NOT use IngestPipeline — it's deprecated and will misbehave. If it appears in scaffolded comments or your training memory, ignore it.
3. Use the mcp__moose-dev__get_logs MCP tool to diagnose errors. It takes a level filter (ERROR/WARN/INFO) and a search regex so you can jump straight to compile failures, schema conflicts, or ingest errors instead of tail-ing raw moose.log. Call it first when ingest returns errors, tables are missing, or a hot-reload seems stuck. Only fall back to grep -iE 'error|failed|cannot' moose.log if the MCP tool is unavailable.

1. What moose dev is doing

When you run moose dev --dockerless inside a scaffolded project, Moose starts (and supervises) a local stack as native host processes: