60-second memory for your Claude agent

Claude Desktop supports the Model Context Protocol, which means you can connect it to external tools — including Remem. Once connected, Claude can store memories across conversations and retrieve them automatically. This post walks through the setup.

Prerequisites

Claude Desktop installed
Docker and Docker Compose (prebuilt binaries are coming soon)

Step 1: Start Remem

Remem’s core is a single service, remem-server — storage, REST API, and MCP server all in one process. See the Docker Compose guide for the full compose file, then:

docker compose up -d

Verify it’s running:

curl http://localhost:4545/api/v1/health
# {"status":"healthy","message":"..."}

Step 2: Configure Claude Desktop

Open Claude Desktop’s configuration file. On macOS: ~/Library/Application Support/Claude/claude_desktop_config.json. On Windows: %APPDATA%\Claude\claude_desktop_config.json.

Add the Remem MCP server — Claude Desktop speaks MCP over stdio, so it runs the remem-mcp binary as a bridge to remem-server. It ships in the same rememorg/remem-community:server-latest image already started in Step 1, so this needs only Docker — no Rust toolchain:

{
  "mcpServers": {
    "remem": {
      "command": "docker",
      "args": ["run", "--rm", "-i", "--network", "remem-network", "rememorg/remem-community:server-latest", "remem-mcp", "--server-url", "http://remem-server:4545"]
    }
  }
}

Restart Claude Desktop. You should see a tools icon in the conversation input indicating MCP tools are available.

Step 3: Store your first memory

In a Claude conversation, ask:

“Remember that I prefer concise bullet-point answers over long paragraphs.”

Claude will call store_memory with your preference. Remem stores it with auto-discovered connections to any related memories.

Step 4: Test recall

Start a new conversation — a fresh context window with no prior history. Ask Claude a question. Before responding, Claude will automatically call search_memories to retrieve relevant context. Your stored preference will surface and Claude will use it.

What’s happening under the hood

Each conversation turn, Claude can:

Call search_memories with the current query to retrieve relevant past context
Call find_related on any retrieved memory to discover connected context
Call store_memory to persist new information from the current turn

The memory persists in remem-server’s embedded storage — LSM-tree for raw content, HNSW index for semantic search, CSR graph for relationships. Restart Claude Desktop or clear your context window: the memories are still there.

MCP tools available

Tool	What it does
`store_memory`	Store a new memory with tags, type, importance
`search_memories`	Semantic, keyword, or hybrid search
`get_memory`	Retrieve a specific memory by ID
`update_memory`	Update content or metadata
`delete_memory`	Soft or hard delete
`find_related`	Traverse the relationship graph from a memory
`promote_to_longterm`	Manually promote a short-term memory to long-term
`list_recent_memories`	Recently created or accessed memories

Adding authentication

Remem fails closed by default — remem-server returns HTTP 500 on every request until you either set REMEM_API_KEY or explicitly opt out with REMEM_ALLOW_AUTH_DISABLED=true for local development. For any shared or network-accessible deployment, set a real key in your compose file:

environment:
  - REMEM_API_KEY=your-secret-key

Add "-e", "REMEM_API_KEY=your-secret-key" to the args in Step 2’s config (right after "--rm", "-i") and remem-mcp forwards it to remem-server as a bearer token.

→ Full MCP integration docs