An engineering organization
Giving agents safe, governed access to production context
role: Designed and built the MCP server and provisioned the shared model infrastructure.
// problem
Agents are only as useful as the context they can reach. Handing them API keys or broad permissions to get that context is not acceptable in production.
// approach
- An MCP server that exposes API data and operations tools to agents over two transports.
- Per-environment Terraform and CI/CD for the server, like any other production service.
- Shared model infrastructure for internal agents on a cloud AI platform, provisioned in Terraform: model deployments, quota, workload identities and least-privilege RBAC.
// architecture
Coding agents
MCP server
two transports
API data
Operations tools
Terraform
per environment
Model deployments + quota
Workload identities + RBAC
Internal agents
// guardrails and production concerns
- Workload identities instead of stored keys.
- Least-privilege RBAC on model access.
- Everything provisioned in Terraform, per environment.
// outcome
Agents get real context without leaked keys or over-broad permissions. AI is treated as infrastructure, with the same rigor as the rest of the platform.
// stack
- Model Context Protocol
- Terraform
- Azure OpenAI
- Workload identity
- RBAC
- CI/CD