Skip to content

MCP Integration

Varun Pratap Bhardwaj edited this page Mar 16, 2026 · 1 revision

MCP Integration

SuperLocalMemory v2.1.0 includes a universal mcp-server that provides native integration with 17+ IDEs including Claude Desktop, Cursor IDE, Windsurf, and more. This local-first, mcp-protocol based integration enables AI assistants to naturally access your memory system without manual commands.

Keywords: mcp-server, claude-desktop, cursor-ide, windsurf, universal-integration, model-context-protocol


🌐 What is MCP?

MCP (Model Context Protocol) is Anthropic's open protocol that allows AI assistants to connect to external tools and data sources. SuperLocalMemory's MCP server provides:

  • 12 Tools - AI-accessible functions (remember, recall, search, status, learning, etc.)
  • 6 Resources - Data feeds (graph, patterns, recent memories, learning status)
  • 2 Prompts - Context injection templates
  • 100% Local - No cloud dependencies, runs entirely on your machine

Why MCP Matters

With MCP, you can say:

You: "Remember that we use FastAPI for all REST APIs"
Claude: [Automatically uses the remember tool] ✓ Saved to memory

No slash commands needed. The AI assistant naturally uses your memory system.

How MCP Works in Practice

SuperLocalMemory runs an MCP server locally on your machine. When you open Claude Desktop, Cursor, or any supported IDE, it automatically connects to this server. Every time you ask your AI assistant to remember something or search your memory, it's making a direct local call — no cloud, no delays, complete privacy.

MCP Protocol Diagram Figure: MCP client connected to SuperLocalMemory, showing protocol and trust score


🚀 Supported IDEs

Auto-Configured During Installation

These IDEs are automatically detected and configured when you run ./install.sh:

IDE Config File Status
Claude Desktop ~/Library/Application Support/Claude/claude_desktop_config.json ✅ Auto
Cursor ~/.cursor/mcp_settings.json ✅ Auto
Windsurf ~/.windsurf/mcp_settings.json ✅ Auto
Continue.dev .continue/config.yaml ✅ Auto

Manual Setup Required

These IDEs require manual configuration (see detailed steps below):

IDE Manual Setup Documentation
ChatGPT Desktop Settings → MCP Setup Guide
Perplexity AI App Settings Setup Guide
Zed Editor JSON config Setup Guide
OpenCode MCP config Setup Guide
Antigravity IDE Settings Setup Guide
Cody Custom commands Setup Guide
Aider Smart wrapper Setup Guide

Total: 17+ supported tools


⚡ Quick Start (Auto-Installation)

Step 1: Install SuperLocalMemory

git clone https://github.com/varun369/SuperLocalMemoryV2.git
cd SuperLocalMemoryV2
./install.sh

The installer will:

  1. Detect installed IDEs (Cursor, Windsurf, Claude Desktop, Continue)
  2. Install MCP Python package if not present
  3. Create MCP configuration files automatically
  4. Configure each detected IDE

Step 2: Verify Installation

Claude Desktop:

  1. Restart Claude Desktop completely
  2. Look for "superlocalmemory-v2" in available tools
  3. Test: "Save this to memory: Test memory"

Cursor:

  1. Restart Cursor IDE
  2. Open AI assistant panel
  3. Test: "Remember that we prefer TypeScript"

Windsurf:

  1. Restart Windsurf
  2. Check MCP tools list
  3. Test: "What's in my recent memories?"

Continue.dev:

  1. Reload VS Code window (Cmd+Shift+P → "Reload Window")
  2. Open Continue panel
  3. Test: Type /slm-remember to see the skill

🛠️ Manual Setup Guides

ChatGPT Desktop

Platform: macOS, Windows

Steps:

  1. Open ChatGPT Desktop App
  2. Go to SettingsMCP Servers
  3. Click Add Server
  4. Enter configuration:
{
  "name": "superlocalmemory-v2",
  "command": "python3",
  "args": [
    "/Users/yourusername/.claude-memory/mcp_server.py"
  ],
  "cwd": "/Users/yourusername/.claude-memory",
  "env": {
    "PYTHONPATH": "/Users/yourusername/.claude-memory"
  }
}
  1. Replace /Users/yourusername with your actual home directory
  2. Click Save and restart ChatGPT

Verify: Ask ChatGPT "What tools do you have access to?" - should see SuperLocalMemory tools.


Perplexity AI

Platform: macOS, Web

Steps:

  1. Open Perplexity App
  2. Navigate to SettingsIntegrationsMCP
  3. Click Add MCP Server
  4. Paste configuration:
{
  "superlocalmemory-v2": {
    "command": "python3",
    "args": [
      "/Users/yourusername/.claude-memory/mcp_server.py"
    ],
    "cwd": "/Users/yourusername/.claude-memory"
  }
}
  1. Save and restart Perplexity

Verify: Ask "Do you have memory tools?" - should list SuperLocalMemory.


Zed Editor

Platform: macOS, Linux

Config File: ~/.config/zed/settings.json

Steps:

  1. Open ~/.config/zed/settings.json in a text editor
  2. Add MCP configuration:
{
  "mcp": {
    "servers": {
      "superlocalmemory-v2": {
        "command": "python3",
        "args": [
          "/Users/yourusername/.claude-memory/mcp_server.py"
        ],
        "cwd": "/Users/yourusername/.claude-memory",
        "env": {
          "PYTHONPATH": "/Users/yourusername/.claude-memory"
        }
      }
    }
  }
}
  1. Save and restart Zed
  2. Open AI assistant panel

Verify: Check available tools in Zed's AI panel.


OpenCode

Platform: Cross-platform

Config File: ~/.opencode/mcp.json

Steps:

  1. Create or edit ~/.opencode/mcp.json:
{
  "mcpServers": {
    "superlocalmemory-v2": {
      "command": "python3",
      "args": [
        "/Users/yourusername/.claude-memory/mcp_server.py"
      ],
      "cwd": "/Users/yourusername/.claude-memory"
    }
  }
}
  1. Restart OpenCode
  2. Check MCP tools list

Antigravity IDE

Platform: Cross-platform

Steps:

  1. Open Antigravity settings

  2. Navigate to ExtensionsMCP

  3. Add server with Python command:

    • Command: python3
    • Script: /Users/yourusername/.claude-memory/mcp_server.py
    • Working Dir: /Users/yourusername/.claude-memory
  4. Apply and restart


Cody

Note: Cody uses custom commands instead of MCP. These are auto-configured by install-skills.sh.

Manual Setup:

  1. Edit .vscode/settings.json (VS Code) or Cody settings (JetBrains):
{
  "cody.customCommands": {
    "slm-remember": {
      "description": "Save to SuperLocalMemory",
      "prompt": "Execute: python3 ~/.claude-memory/skills/slm-remember/main.py \"${input}\""
    },
    "slm-recall": {
      "description": "Search SuperLocalMemory",
      "prompt": "Execute: python3 ~/.claude-memory/skills/slm-recall/main.py \"${input}\""
    }
  }
}
  1. Reload VS Code or JetBrains IDE

Usage: Type /slm-remember in Cody panel.


Aider

Note: Aider uses the smart wrapper instead of MCP.

Setup:

  1. Run the install script (already done if you installed SuperLocalMemory)
  2. Use aider-smart instead of aider:
# Instead of: aider myfile.py
# Use:
aider-smart myfile.py

What it does:

  • Automatically injects recent memories as context
  • Uses standard slm CLI commands
  • No configuration needed

🔧 MCP Server Tools

12 Available Tools

Tool Purpose Example AI Request
remember() Save memory "Remember that we use FastAPI"
recall() Search memories "What do we use for REST APIs?"
list_recent() Show recent memories "Show my recent memories"
get_status() System statistics "What's my memory system status?"
build_graph() Rebuild knowledge graph "Rebuild my knowledge graph"
switch_profile() Change profile "Switch to work profile"
search() Search memories (OpenAI MCP spec) Used by ChatGPT Connectors
fetch() Fetch memory by ID (OpenAI MCP spec) Used by ChatGPT Connectors
backup_status() Auto-backup status "What's the backup status?"
memory_used() Feedback for learning (v2.7) Implicit when recalled memory is used
get_learned_patterns() Retrieve learned patterns (v2.7) "What patterns have you learned?"
correct_pattern() Correct a learned pattern (v2.7) "I prefer Vue, not React"

6 Resources

Resource Content Access
memory://graph/clusters Knowledge graph clusters Read-only
memory://patterns/identity Learned patterns Read-only
memory://recent/{limit} Recent memories feed Read-only
memory://stats System statistics Read-only
memory://learning/status Learning system status (v2.7) Read-only
memory://engagement Engagement metrics (v2.7) Read-only

2 Prompts

  1. Context Injection - Automatically provides recent memories to AI
  2. Identity Profile - Injects coding preferences from pattern learning

✅ Verify MCP Connection

Once you've configured MCP in your IDE, check the dashboard to confirm the connection is active:

# Start the web dashboard
python3 ~/.claude-memory/ui_server.py

# Dashboard opens at http://localhost:8765
# Navigate to the "Agents" tab

You should see your IDE listed as a connected MCP client:

MCP Agents Connected Figure: Dashboard showing MCP client connected with trust score 1.00

Key indicators:

  • Protocol: Shows MCP (agent-to-tool)
  • Trust Score: Starts at 1.00 (full trust)
  • Last Seen: Recent timestamp means active connection
  • Memory Operations: Count of remember/recall calls this session

Note: The Agents tab (v2.5 feature) tracks all MCP connections in real-time. This view helps you verify that your IDE is successfully connected and communicating with SuperLocalMemory.


🐛 Troubleshooting

MCP Server Not Showing Up

Problem: IDE doesn't list SuperLocalMemory tools

Solutions:

  1. Check Python path:

    which python3
    # Output: /usr/bin/python3 or /opt/homebrew/bin/python3

    Update config file with correct path.

  2. Verify MCP package:

    pip3 install mcp
  3. Check install path:

    ls ~/.claude-memory/mcp_server.py
    # Should exist
  4. Restart IDE completely - Quit and reopen, don't just reload.

  5. Check logs:

    • Claude Desktop: ~/Library/Logs/Claude/mcp.log
    • Cursor: ~/.cursor/logs/mcp.log

"Permission Denied" Error

Solution: Make script executable:

chmod +x ~/.claude-memory/mcp_server.py

"Module not found" Error

Solution: Install required packages:

cd ~/.claude-memory
pip3 install -r requirements.txt

MCP Server Crashes

Solution: Check Python version:

python3 --version
# Must be 3.8+

Configuration File Not Found

Solution: Create directory:

# Claude Desktop (macOS)
mkdir -p ~/Library/Application\ Support/Claude

# Cursor (macOS)
mkdir -p ~/.cursor

# Windsurf (macOS)
mkdir -p ~/.windsurf

[More troubleshooting: MCP-TROUBLESHOOTING.md →]


📖 Complete Manual Setup Guide

For detailed manual setup instructions for all IDEs, see:

docs/MCP-MANUAL-SETUP.md

Includes:

  • Step-by-step configuration for each IDE
  • Platform-specific paths (macOS/Windows/Linux)
  • Troubleshooting for each tool
  • Custom MCP client examples

🔗 Related Pages


Created by Varun Pratap Bhardwaj

Clone this wiki locally