- apps/runner/: Dockerfile (node22 + python3 + uv + bun) + supervisor.mjs
(HTTP API for load/call/unload/health)
- docker-compose: runner service with shared services-data volume
- Broker mcp_deploy: git clone or zip extract → runner /load → MCP spawn
- Broker mcp_call: routes managed services to runner via HTTP, falls back
to live-proxy for peer-hosted servers
- RUNNER_URL env var for broker → runner communication
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Files: MinIO-backed file sharing built into the broker.
share_file for persistent mesh files, send_message(file:) for
ephemeral attachments. Presigned URLs for download, access
tracking per peer.
Broker infra: MinIO in docker-compose, internal network.
HTTP POST /upload endpoint. WS handlers for get_file,
list_files, file_status, delete_file.
Multi-target: send_message(to:) accepts string or array.
Targets deduplicated before delivery.
Targeted views: MCP instructions teach Claude to send
tailored messages per audience instead of generic broadcasts.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
One-shot migrate container runs drizzle-kit migrate against DATABASE_URL
and exits 0 before web boots. web service depends_on with condition
service_completed_successfully, so failed migrations block web startup
instead of serving 500s against a stale schema. Broker deliberately does
NOT depend on migrate - it tolerates DB-down gracefully per DEPLOY_SPEC
and should keep serving WS peers even during migration failures.
Also excludes apps/cli from docker build context (CLI ships to npm, not
containers) to sidestep zod spec drift in its package.json vs lockfile.
Known followup: migrate image is 3.27GB due to pnpm catalog: specifiers
forcing full-workspace resolution. pnpm deploy bundle trim is a P2.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
