solon

Solon MCP Connector Security Hardening Guide

Least-privilege permissions

• Grant agents only the specific API access they need – Don't give the AI agent unrestricted access to your entire system. Limit it to read-only for sensitive data, and write-only to designated sandboxed areas (like temporary work folders).
• Isolate secrets storage – Keep API keys, database passwords, and enterprise credentials in a separate secure vault (like HashiCorp Vault or AWS Secrets Manager), not embedded in the agent's configuration. Agents should request secrets through a controlled gatekeeper, not store them directly.
• Restrict database and file system permissions – Ensure the Solon agent process runs as a low-privilege user account that cannot access production data, payroll systems, or confidential files without explicit approval per request.
• Rotate credentials frequently – Any credentials the agent uses should expire and be replaced every 30–90 days to limit damage if they're compromised.

Network isolation

• Run agents in a dedicated network zone – Deploy the Solon connector in a separate, segmented network (DMZ or private subnet) so it can't directly reach your core systems; require all communication through monitored gateways.
• Whitelist outbound connections only – Configure firewall rules so the agent can only connect to pre-approved external services (e.g., specific APIs); block any unexpected outbound traffic.
• Disable direct agent-to-agent communication – If you run multiple AI agents, prevent them from talking to each other directly; route all inter-agent traffic through a central audit point.
• Use encrypted channels (TLS/HTTPS) – Ensure all data traveling between the agent and your systems is encrypted in transit; never send unencrypted credentials or sensitive data over the network.

Monitoring

• Log all agent actions – Record every command the agent executes, every file it accesses, and every API call it makes in a central, tamper-proof log (separate from systems the agent can modify).
• Set up real-time alerts for unusual behavior – Create alerts for unexpected large data transfers, access to restricted areas, failed credential attempts, or agents connecting to unapproved external services.
• Track dependency updates and security patches – Since Solon is a smaller project than Spring, manually check for published security advisories and patches at least weekly; subscribe to GitHub release notifications.
• Audit third-party integrations – Regularly review which external tools, libraries, and plugins the Solon agent uses; remove unused ones and verify that active ones still receive security updates.

Red flags to watch for

• Credentials appearing in agent logs or error messages – This indicates improper secret handling; immediately rotate affected credentials and investigate where they leaked.
• Unexpected file creation or modification outside designated work areas – This may signal the agent has been compromised or misconfigured; halt the agent and review its recent activity.
• Delays or gaps in security patch releases from the Solon project – If the maintainers stop issuing updates for 3+ months, escalate to leadership