Claude Desktop Integration

Give Claude Desktop persistent, encrypted memory in under 5 minutes. Every memory passes through PII sanitization, intent validation, and AES-256-GCM encryption before storage.

On this page

Overview
Prerequisites
Configuration
Usage
Available Tools
Troubleshooting

Overview

Claude Desktop supports the Model Context Protocol (MCP), which lets you connect external tools as first-class capabilities. By registering Synapse Layer as an MCP server, Claude gains two primary tools — store_memory and recall_memory — plus 11 additional specialized tools for memory management, health checks, and context initialization.

All memory operations go through the Cognitive Security Pipeline: PII detection and redaction, intent classification, differential privacy noise injection, and AES-256-GCM encryption at rest.

Prerequisites

Claude Desktop installed (macOS or Windows)
Python 3.10+
Synapse Layer package installed: pip install synapse-layer "mcp[cli]"

Configuration

Add Synapse Layer to your claude_desktop_config.json:

json

{
  "mcpServers": {
    "synapse-layer": {
      "command": "uvx",
      "args": [
        "--from",
        "synapse-layer",
        "synapse-layer"
      ]
    }
  }
}

Note

The uvx launcher auto-installs the package in an isolated environment. Alternatively, use npx with the Smithery registry:

json

{
  "mcpServers": {
    "synapse-layer": {
      "command": "npx",
      "args": [
        "-y",
        "@smithery/cli@latest",
        "run",
        "@anthropics/synapse-layer",
        "--config",
        "{}",
        "--key",
        "YOUR_SMITHERY_KEY"
      ]
    }
  }
}

For environment configuration, create a .env file in the working directory:

bash

# Unique identifier for this agent's memory namespace
SYNAPSE_AGENT_ID=claude-desktop-agent

# Differential-privacy epsilon (0.1 = max privacy, 1.0 = balanced)
SYNAPSE_PRIVACY_EPSILON=0.5

# Logging level: DEBUG | INFO | WARNING | ERROR
LOG_LEVEL=INFO

Usage

Once configured, Claude Desktop automatically discovers the Synapse Layer tools. You can ask Claude to use memory naturally:

You:"Remember that I prefer TypeScript and my project uses Next.js 14."

Claude:I've stored that preference. The memory was processed through the security pipeline — PII sanitization applied, intent classified as "preference" with 0.92 confidence, and encrypted with AES-256-GCM.

You:"What do you know about my project setup?"

Claude:Based on my memory, you prefer TypeScript and your project uses Next.js 14. (Trust Quotient: 0.95)

Available Tools

When connected, Claude Desktop has access to all 13 Synapse Layer MCP tools. The most commonly used in interactive sessions:

Tool	Description
store_memory	Store text through the full security pipeline
recall_memory	Semantic search across stored memories
list_memories	Browse all memories for this agent
health_check	Verify pipeline status and version
initialize_context	Bootstrap agent context from existing memories

See the MCP Tools Reference for the complete list of all 13 tools with parameter schemas.

Troubleshooting

Tools not appearing in Claude Desktop

Verify the config path is correct and restart Claude Desktop. On macOS: ~/Library/Application Support/Claude/claude_desktop_config.json

Connection errors

Ensure Python 3.10+ is in your PATH. Run uvx --from synapse-layer synapse-layer manually to check for dependency issues.

Memories not persisting between sessions

Ensure you have set a consistent SYNAPSE_AGENT_ID in your environment. Different agent IDs create separate memory namespaces.

Next Steps

Cursor

Use Synapse Layer with Cursor IDE.

MCP Tools Reference

Full reference for all 13 MCP tools.

MCP Server Setup

Advanced MCP server configuration.