Claude Desktop Integration Guide

Give Claude Desktop persistent memory across conversations using Engram.

Setup

1. Get an API Key

2. Add the MCP Server

Edit your Claude Desktop config file:

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


{
  "mcpServers": {
    "engram": {
      "url": "https://mcp.getengram.app/mcp",
      "headers": {
        "Authorization": "Bearer engram_sk_live_your_key_here"
      }
    }
  }
}

3. Restart Claude Desktop

Quit and reopen Claude Desktop. Engram’s 6 tools will appear in Claude’s tool list (click the hammer icon to verify).

Automatic Memory

Claude Desktop supports Projects with custom instructions. This is how you set up auto-memory.

Create a Project with Memory Instructions

Open Claude Desktop
Click Projects in the sidebar
Create a new project (or open an existing one)
Click the project settings and add Custom Instructions:


You have access to Engram memory tools. Use them automatically — the user should not need to ask you to remember or recall.

## On conversation start

Search Engram for context relevant to the user's first message:

  search
    query: "<summary of what the user is asking about>"
    limit: 5

If results are relevant, use them to inform your response. Mention if you found relevant prior context ("I recall from our previous conversation that...").

## During the conversation

When you learn something important, store it:

  create_conversation
    title: "<concise description>"
    agent_id: "claude-desktop"
    tags: ["<topic>"]

  append_messages
    conversation_id: "<id>"
    messages:
      - role: "user"
        content: "<what the user said>"
      - role: "assistant"
        content: "<what you said and why>"

## What to store

- User preferences and personal details they share
- Decisions and reasoning
- Important facts the user wants remembered
- Project context and goals
- Anything the user would expect you to know next time

## What NOT to store

- Casual greetings or small talk
- Information you can look up
- Temporary or time-sensitive details

How It Works

Every conversation in that project uses the custom instructions
Claude searches Engram at the start of each conversation
Important context is stored during the conversation
Next conversation, Claude recalls relevant prior context

Example: Memory in Action

Conversation 1:


You: I'm starting a new project using Next.js and Supabase.
     We need to support multi-tenancy with row-level security.

Claude: [stores in Engram:
  title: "New project: Next.js + Supabase with multi-tenancy"
  tags: ["project-setup", "architecture"]
  messages: project stack choice and multi-tenancy requirement]

Claude: Great choice. For multi-tenancy with Supabase RLS, I'd recommend...

Conversation 2 (days later):


You: How should we handle the auth flow?

Claude: [searches Engram → finds conversation about Next.js + Supabase project]
Claude: Since we're using Next.js with Supabase and need multi-tenancy
        with RLS, I'd recommend using Supabase Auth with custom claims
        for the tenant ID. This way the RLS policies can reference
        auth.jwt()->'tenant_id' directly...

Claude remembers the project context without being reminded.

Multiple Projects

You can create separate projects for different areas of your life:

Project	Use case
Work - Engineering	Technical decisions, codebase context
Work - Management	Meeting notes, team context
Personal	Preferences, travel plans, recommendations
Research	Topics you’re exploring, papers discussed

All projects share the same Engram memory (same API key), so context from one project can surface in another if relevant. The agent_id and tags help distinguish where memories came from.

Tips

Be specific in project instructions. The more context you give Claude about what to remember, the better its memory will be.
Use different tags per project. This helps filter memories when searching.
Mention prior context. If Claude recalls something from Engram, it will mention it. If it misses something, just say “remember when we discussed X?” — it will search again.
Works with Claude Code. If you use the same API key in Claude Code and Claude Desktop, they share memory. Debug a bug in Claude Code, discuss the architecture in Claude Desktop.