Skip to main content
This guide walks you through installing AIP and running your first secured MCP server.

​
Prerequisites

Before installing AIP, ensure you have:
  • Go 1.21+ (for building from source)
  • Python 3.11+ or Node.js 20+ (for MCP servers)
  • An MCP-compatible AI client (Cursor IDE, Claude Desktop, VS Code)

​
Installation Methods

Install directly from the Go implementation repository:
go install github.com/openagentidentityprotocol/aip-go/cmd/aip@latest
Verify installation:
aip --version

​
Quick Start

1

Create a policy file

Create agent.yaml with a simple read-only policy:
apiVersion: aip.io/v1alpha1
kind: AgentPolicy
metadata:
  name: secure-agent
spec:
  mode: enforce
  allowed_tools:
    - read_file
    - list_directory
    - git_status
  tool_rules:
    - tool: write_file
      action: ask        # Human approval required
    - tool: exec_command
      action: block      # Never allowed
2

Start the AIP proxy

Wrap your existing MCP server:
# For a Python MCP server
aip --target "python mcp_server.py" --policy ./agent.yaml

# For a Node.js MCP server
aip --target "npx @mcp/server" --policy ./agent.yaml
The proxy will start on stdio and forward to your MCP server.
3

Configure your AI client

Generate configuration for your AI client:
# For Cursor IDE
aip --generate-cursor-config --policy ./agent.yaml --target "npx @mcp/server"
This outputs JSON configuration to add to your client’s MCP settings.
4

Test the policy

Try using a blocked tool in your AI client:
User: "Run the command 'rm -rf /'"
Agent: [attempts exec_command]
AIP: ❌ Permission Denied (logged to audit trail)

​
Verification

​
Check Policy Evaluation

Verify the proxy is enforcing policy: 
# Start in monitor mode to see what would be blocked
aip --target "python mcp_server.py" --policy ./agent.yaml --mode monitor
Monitor mode logs violations but allows them through - useful for testing.

​
View Audit Logs

Audit logs are written to aip-audit.jsonl by default: 
cat aip-audit.jsonl | jq
Example log entry: 
{
  "timestamp": "2026-03-03T16:32:00Z",
  "agent_id": "agent-abc123",
  "user": "[email protected]",
  "tool": "exec_command",
  "arguments": {"cmd": "rm -rf /"},
  "decision": "deny",
  "reason": "tool not in allowed_tools"
}

​
System Requirements

​
Performance

  • Memory: 10-20 MB per proxy instance
  • CPU: Minimal (less than 1% idle)
  • Disk: Audit logs grow at approximately 1KB per request

​
Supported Platforms

  • Linux: All distributions (amd64, arm64)
  • macOS: 10.15+ (Intel and Apple Silicon)
  • Windows: Windows 10+ (experimental)

​
Troubleshooting

If you see validation errors on startup:
# Validate your policy file
aip --validate-policy ./agent.yaml
Common issues:
  • Invalid YAML syntax
  • Unknown fields in spec
  • Regex compilation errors in allow_args
If the proxy can’t connect to your MCP server:
  • Verify the target command works standalone: python mcp_server.py
  • Check stdio/stderr output for errors
  • Ensure the MCP server implements the protocol correctly
For action: ask to work:
  • macOS: Requires terminal to have accessibility permissions
  • Linux: Requires zenity or kdialog installed
  • Windows: Not yet supported

​
Next Steps

Writing Policies

Learn to write comprehensive agent policies

DLP Configuration

Configure data loss prevention scanning

Deployment

Deploy AIP in production environments

Audit Logging

Set up compliance-ready audit trails

Build docs developers (and LLMs) love

Get started for freeTalk to us