How do I give Claude Desktop persistent memory?

Add Dakera's MCP server configuration to your Claude Desktop settings. Claude then has access to 83 memory tools for storing and recalling information across conversations.

Does Claude Desktop memory work across conversations?

Yes. With Dakera, memories stored in one Claude conversation are available in all future conversations. Claude can recall user preferences, past decisions, and project context.

Is my Claude Desktop memory data private?

Yes. Dakera runs entirely on your local machine. No data leaves your computer — memories are stored locally with AES-256-GCM encryption at rest.

CLAUDE DESKTOP · MCP

Claude Desktop Integration

Give Claude Desktop persistent memory that survives across conversations. Dakera's MCP server connects directly — Claude remembers your preferences, project context, and past interactions without re-explaining every session.

Native MCP support. Claude Desktop supports MCP servers natively. Add one config block to claude_desktop_config.json and you're set. All memory stays on your infrastructure.

Quick Start

Start Dakera

docker run -d --name dakera \
  -p 3300:3300 \
  -e DAKERA_API_KEY=dk-your-key-here \
  -v dakera-data:/data \
  ghcr.io/dakera-ai/dakera:latest

Configure Claude Desktop

Edit claude_desktop_config.json:

macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "dakera": {
      "command": "docker",
      "args": ["run", "--rm", "-i",
        "--network", "host",
        "ghcr.io/dakera-ai/dakera-mcp:latest"],
      "env": {
        "DAKERA_URL": "http://localhost:3300",
        "DAKERA_API_KEY": "dk-your-key-here",
        "DAKERA_AGENT_ID": "claude-desktop"
      }
    }
  }
}

Restart Claude Desktop

Quit and reopen Claude Desktop. You'll see a hammer icon indicating MCP tools are available. Claude now has access to 83 memory tools.

What Claude Can Do with Memory

Capability	What It Means
Store memories	Save facts, preferences, and context with importance scores
Recall by meaning	Hybrid vector + keyword search finds relevant past context
Session tracking	Group related memories by conversation session
Knowledge graphs	Build entity relationships from conversations (people, projects, decisions)
Memory decay	Old, unaccessed memories fade naturally — keeping recall sharp
Namespace isolation	Separate work memory from personal memory

Example Interactions

Personal Preferences

Claude automatically stores preferences you mention:

You: "I prefer TypeScript over JavaScript and always use strict mode"
Claude: [stores with importance 0.8, tags: preferences, coding]

// Next session:
You: "Help me set up a new project"
Claude: [recalls preference] "I'll set this up with TypeScript in strict mode,
since that's your preference."

Project Context

You: "We're building a SaaS with Next.js, Prisma, and Stripe"
Claude: [stores project architecture context]

// Days later:
You: "Add a webhook handler for subscription changes"
Claude: [recalls Stripe + Next.js context] "I'll create a Next.js API route
at /api/webhooks/stripe that handles subscription.updated events using
your Prisma schema..."

Research & Notes

You: "I've been researching RAG approaches. Key finding: hybrid retrieval
     outperforms pure vector by 15-20% on factual recall tasks."
Claude: [stores as high-importance research finding]

// Later:
You: "Should we use pure vector search for our docs chatbot?"
Claude: [recalls research] "Based on your earlier research, hybrid retrieval
outperforms pure vector by 15-20% on factual recall. I'd recommend combining
vector with BM25..."

Privacy & Data Control

Self-hosted: All memory data stays on your machine or server. Nothing leaves your infrastructure.
Encrypted: AES-256-GCM encryption at rest. Even if someone accesses the data files, they can't read your memories.
You own the data: Export, backup, or delete any time. No lock-in.
Per-namespace control: Create separate namespaces for work, personal, or different projects.

Configuration Options

Variable	Description	Default
DAKERA_URL	Dakera server address	http://localhost:3300
DAKERA_API_KEY	Authentication key	Required
DAKERA_AGENT_ID	Unique identifier for this agent	"default"
DAKERA_NAMESPACE	Memory namespace for isolation	"default"

Works with Claude Pro and Team plans. Any Claude Desktop version with MCP support works with Dakera. No additional subscription needed beyond Claude itself.

Next Steps

Full MCP Server documentation — all 83 tools explained
Quick Start guide — detailed server setup
Cursor integration — same memory for AI coding
Memory Patterns — common use cases and architectures