@onenomad/przm-cortex

Cortex MCP server + CLI. Loads config, wires up providers and adapters, schedules ingestion, and exposes work-specific tools to AI clients over MCP.

Ships the cortex bin:

cortex init    # interactive setup wizard (run this first)
cortex start   # boot the MCP server over stdio
cortex smoke   # live probe of every enabled LLM provider
cortex help    # usage

Structure

• src/cli/ — CLI entry and subcommands (init, start, smoke).
• src/mcp/server.ts — MCP server. Zero tools in Phase 1.
• src/mcp/tools/ — one file per tool (filled in starting Phase 2).
• src/clients/engram.ts — typed wrapper over Engram MCP.
• src/clients/persona.ts — typed wrapper over Persona MCP.
• src/registry/providers.ts — loads enabled LLM providers, builds the router.
• src/registry/adapters.ts — loads enabled source adapters (stub).
• src/scheduler.ts — runs adapters on cron (stub).
• src/config.ts — loads and validates config/cortex.yaml.
• src/logger.ts — stderr-only structured logger (stdout is reserved for MCP).

Running

pnpm --filter @onenomad/przm-cortex dev    # watch mode, runs `cortex start`
pnpm --filter @onenomad/przm-cortex build && pnpm --filter @onenomad/przm-cortex start

Tests:

pnpm --filter @onenomad/przm-cortex test       # unit
pnpm --filter @onenomad/przm-cortex test:int   # integration (requires live Ollama)