How to Deploy MCP Memory Service with Docker: Step-by-Step Tutorial

In this hands-on tutorial you'll deploy MCP Memory Service with Docker, connect it to a Claude-style MCP client, and verify that your agent can actually remember facts across sessions. All you need is a machine with Docker and a few minutes.

Step 1 — Pull and run the container

The official image doobidoo/mcp-memory-service:latest is on Docker Hub with over 103K pulls. Start it with persistent storage:

docker run -d --name mcp-memory \
  -p 8080:8080 \
  -v ./data/mcp-memory:/data \
  doobidoo/mcp-memory-service:latest

The volume keeps your knowledge graph intact across container restarts and upgrades.

Step 2 — Health check

curl http://localhost:8080/health

A 200 OK response means the REST API and memory engine are up.

Step 3 — Connect an MCP client

Point Claude Desktop, OpenCode or any MCP client at the server. In your client config register the endpoint, then ask your agent to store a fact — e.g. "Remember that the production database is Postgres 16". Start a new session and ask what it knows about the production stack: the fact is retrieved from the knowledge graph, not the conversation.

Step 4 — Verify persistence

Production tips

• Use docker compose up -d with a restart policy for resilience.
• Allocate at least 2 vCPU / 4 GB RAM; 4 vCPU / 8 GB recommended for heavy pipelines.
• Back up the /data volume — it holds your agents' entire memory.

That's it — your agents now have durable memory. From here, wire the REST API into LangGraph or CrewAI pipelines and watch your multi-step workflows stop repeating themselves.