AgentMesh for Claude Desktop

April 26, 2026 · View on GitHub

The "Day 0" Integration: Make every Claude user realize their local tool use is unsafe, then fix it with one command.

The Problem

When you connect Claude Desktop to a local MCP server (like the Filesystem server), Claude has unfettered access to your files. One hallucination could:

  • Delete critical files
  • Expose sensitive data
  • Modify system configurations
  • Access credentials

Claude Desktop has no governance layer by default.

The Solution: AgentMesh MCP Proxy

AgentMesh provides a transparent proxy that sits between Claude Desktop and any MCP server, adding:

  • 🔒 Policy Enforcement - Block dangerous operations before they happen
  • 📊 Trust Scoring - Continuous behavioral monitoring
  • 📝 Audit Logging - Tamper-evident logs of every action
  • ✅ Verification Footers - Visual confirmation that governance is active

Quick Start

1. Install AgentMesh

pip install agentmesh-platform

2. Set Up Claude Desktop Integration

agentmesh init-integration --claude

This command will:

  • Locate your claude_desktop_config.json
  • Backup the existing configuration
  • Add an example AgentMesh-protected filesystem server
  • Provide next steps

3. Restart Claude Desktop

After restarting, Claude will connect to MCP servers through AgentMesh.

Manual Configuration

If you prefer to configure manually, edit your claude_desktop_config.json:

Before (Unsafe)

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "/Users/me"]
    }
  }
}

After (Protected by AgentMesh)

{
  "mcpServers": {
    "filesystem": {
      "command": "agentmesh",
      "args": [
        "proxy",
        "--target", "npx",
        "--target", "-y",
        "--target", "@modelcontextprotocol/server-filesystem",
        "--target", "/Users/me"
      ]
    }
  }
}

Policy Levels

AgentMesh supports three policy levels:

Strict (Default)

  • Blocks all write/delete operations
  • Blocks access to sensitive paths (/etc, /root, /.ssh)
  • Allows read operations only
  • Best for: Production environments, shared machines
{
  "args": ["proxy", "--policy", "strict", "--target", "..."]
}

Moderate

  • Warns on write operations but allows them
  • Blocks access to critical system paths
  • Allows most operations with logging
  • Best for: Development environments
{
  "args": ["proxy", "--policy", "moderate", "--target", "..."]
}

Permissive

  • Allows all operations
  • Logs everything for audit
  • Best for: Testing, controlled environments
{
  "args": ["proxy", "--policy", "permissive", "--target", "..."]
}

Verification Footers

Every tool response from Claude will include a verification footer:

> 🔒 Verified by AgentMesh (Trust Score: 980/1000)
> Agent: did:agentmesh:mcp-proxy:abc123...
> Policy: strict | Audit: Enabled

This provides:

  • Visual confirmation that AgentMesh is active
  • Trust score showing behavioral health
  • Audit status for compliance

Disable Footers

If you prefer cleaner output:

{
  "args": ["proxy", "--no-footer", "--target", "..."]
}

Examples

Protect Multiple MCP Servers

{
  "mcpServers": {
    "filesystem": {
      "command": "agentmesh",
      "args": [
        "proxy",
        "--policy", "strict",
        "--target", "npx",
        "--target", "-y",
        "--target", "@modelcontextprotocol/server-filesystem",
        "--target", "/Users/me/safe-directory"
      ]
    },
    "database": {
      "command": "agentmesh",
      "args": [
        "proxy",
        "--policy", "moderate",
        "--target", "python",
        "--target", "my_db_server.py"
      ]
    }
  }
}

Support

"You wouldn't run a website without SSL; don't run an agent without AgentMesh."