Shared services

README

Shared Embedding Daemon (opt-in)

docker-compose.yml in this directory runs the shared mcp-embedding-daemon as a long-lived local container. It is opt-in — each plugin works standalone without it (each loads its own embedding model in-process). The shared daemon exists for power users running multiple plugins that benefit from a single model load + warm cache.

When to use

Enable the shared daemon if any of these apply:

• You run wet + mnemo + crg concurrently and want lower RAM usage.
• Your machine has < 16 GB RAM and you want headroom for IDE / browser.
• You want the embedding model warm across plugin restarts (faster cold start when Claude Code re-spawns a plugin).

Skip the shared daemon if:

• You only run one of the three plugins.
• You have plenty of RAM (≥ 32 GB) and prefer fewer moving parts.
• You cannot run Docker on your machine.

Operations

Stop the daemon

cd ~/projects/mcp-core/docs/shared-services
docker compose down

Plugins fall back to in-process embedding on the next request.

Update the daemon

cd ~/projects/mcp-core/docs/shared-services
docker compose pull
docker compose up -d

View logs

docker compose logs -f embedding-daemon

Cache location

The embedding model + warm cache live in the named volume embedding-cache. Inspect it:

docker volume inspect shared-services_embedding-cache

To wipe it (forces re-download on next start):

docker compose down -v

References

• mcp-core/packages/embedding-daemon/ — source for the daemon itself.
• Migration doc: ~/projects/mcp-core/docs/migration-2026-04-30.md — context for the multi-mode architecture this slot-fits into.