Skip to main content

Infrastructure Independence — Strategic Initiative

Classification: Internal — Strategic Initiative Prepared by: Anshin Health Engineering Version: 2.0, 2026-03-22


Goal

Anshin Health operates its entire infrastructure stack — network, security, DNS, secrets, VPN, user access, Kubernetes, monitoring, certificates, and domains — without relying on any individual consultant or external contractor for routine operations.

North star: An authorized operator opens a terminal, types a question or instruction in plain English, and Claude Code executes it safely — with full audit trail, rollback ability, and documented rationale.


Operational Interface — Claude Code + MCP

All infrastructure operations are performed through Claude Code connected to purpose-built MCP servers. Each MCP server gives Claude direct, audited, role-controlled access to a specific system.

SystemMCP ServerWhat It Manages
Firewall / Networkmcp-opnsenseFirewall rules, VLANs, routing, Suricata IDS
Switch Fabricmcp-omadaTP-Link Omada SDN — ports, LAGs, VLANs
Secretsmcp-infisicalAll credentials, tokens, API keys
Identity / Usersmcp-freeipaUser accounts, groups, Kerberos, PKI
VPN Accessmcp-netbirdPeer management, access policies, groups
Kubernetesmcp-kubectlDeployments, namespaces, services
CI/CDmcp-gitlabPipelines, repositories, merge requests
Monitoringmcp-prometheusMetrics, alerts, dashboards
Config Mgmtmcp-ansibleOS configuration, provisioning
DNS / Domainsmcp-porkbunExternal DNS records, domain management

Access Segregation

Infrastructure access is role-controlled at three levels:

Anshin Staff — Full internal access (all VLANs, K8s, Proxmox/OpenStack)

Onnex Staff — Scoped access (partner VLANs, designated K8s namespaces only)

External Customers — DMZ only (public-facing app endpoints, no internal network access)

Each role maps to: FreeIPA group → NetBird VPN policy → K8s RBAC → SaaS Guard tenant scope.


Implementation Phases

PhaseMilestoneStatus
0MCP servers for identity (FreeIPA) + VPN (NetBird) + DNS (Porkbun)🔵 In Progress
1Network hardware cutover — OPNsense + TP-Link Omada🔵 Planned
2OpenStack deployment on HP#2 + HP#3, migrate off Proxmox📋 Scheduled
3GPU farm activation (HP#4), K8s GPU node pool📋 Scheduled
4Full MCP operational stack — complete AI-assisted administration📋 Scheduled

Governance

All infrastructure changes initiated through Claude Code are:

  • Logged — every MCP operation is timestamped and attributed
  • Reversible — configuration changes are committed to git before application
  • Approved — destructive or security-sensitive operations require explicit Bryan confirmation in the chat interface (Prime Directive)

Operational runbooks and detailed implementation procedures are maintained in the private knowledge store, accessible only to authorized Engineering personnel.


Detailed Runbooks

Full implementation runbooks, MCP configuration details, user provisioning workflows, and network cutover procedures are maintained in the private engineering knowledge store (not published here). Contact the Engineering Director for access.