Episode 13: MCP Integration — Search Dev Memories from Claude Desktop

⏱ Est. reading time: 8 min Updated on 5/7/2026

This episode's scenario: You're discussing the blog project's next steps in Claude Desktop. You say, "What was our database schema design?" But Claude Desktop doesn't know — that was done in Claude Code. If only they could share memories...

13.1 The Problem: Siloed Memories

By default:

Claude Code Claude Desktop
Access to Claude-Mem? ✅ Auto-injected ❌ No access
Can search dev history? ✅ Via MCP tools ❌ No search capability

MCP (Model Context Protocol) bridges this gap.

13.2 What Is MCP?

MCP is Anthropic's Model Context Protocol — a standardized interface letting LLMs call external tools. An MCP Server acts as a "translator" between Claude Desktop's requests and Claude-Mem's database.

13.3 Setup Steps

Step 1: Find the Config File

OS Config Path
macOS ~/Library/Application Support/Claude/claude_desktop_config.json
Windows %APPDATA%\Claude\claude_desktop_config.json
Linux ~/.config/claude/claude_desktop_config.json

Step 2: Add MCP Server Config

{
  "mcpServers": {
    "mcp-search": {
      "command": "node",
      "args": [
        "/Users/YOUR_USERNAME/.claude/plugins/marketplaces/thedotmack/plugin/scripts/mcp-server.cjs"
      ]
    }
  }
}

⚠️ Replace YOUR_USERNAME with your actual system username!

Step 3: Restart Claude Desktop

Fully quit and reopen the application (not just close the conversation).

Step 4: Verify

Ask in Claude Desktop:

List all the MCP tools you can use.

If configured correctly, Claude will list search, timeline, and get_observations.

13.4 Using Memory Search in Claude Desktop

sequenceDiagram
    participant U as You
    participant CD as Claude Desktop
    participant MCP as MCP Server
    participant DB as Claude-Mem DB

    U->>CD: How was our blog DB schema designed?
    CD->>MCP: search(query="database schema", project="my-blog")
    MCP->>DB: FTS5 + vector hybrid query
    DB-->>MCP: Return matching indices
    MCP-->>CD: Return search results
    CD->>MCP: get_observations(ids=[5, 12, 23])
    MCP->>DB: Query by ID
    DB-->>MCP: Return full details
    MCP-->>CD: Return Observation content
    CD-->>U: Your blog DB has Post, Tag, Comment, User, Notification tables...

13.5 Claude Code vs Claude Desktop

Claude Code Search Claude Desktop MCP
Auto context injection ✅ SessionStart ❌ Must ask explicitly
Search capability ✅ Full ✅ Full (via MCP)
Best for Writing code Planning/discussing/reviewing
Requires project dir Yes No

Best practice: Code → Claude Code (auto memory). Discuss/plan → Claude Desktop + MCP.

Hands-On Exercise

  1. Configure Claude Desktop MCP using the steps above
  2. Restart Claude Desktop
  3. Verify: "What MCP tools do you have?"
  4. Ask about the blog project: "What architecture decisions have we made?"
  5. Verify the answer is based on real development memory

Coming Up Next

System is running. But how to fine-tune Claude-Mem's behavior? Switch to English/Chinese mode? Protect sensitive data? Control context injection volume? Next: advanced configuration.

➡️ Episode 14: Advanced Configuration — Customize Your Memory System