Anonymised under NDA
Authenticated MCP server across a production estate
Cross-industry · An online platform operator
A single authenticated Model Context Protocol (MCP) server exposing an operator's infrastructure, web application firewall, logging, and backend APIs to any AI client, so agents can pull data from every platform through one governed interface.
The problem
- Operational data was spread across cloud infrastructure, the web application firewall, centralised logging, and several backend platforms, each with its own console and credentials.
- Every AI experiment started by hand-building a new integration to the same systems, and none of them handled authentication or audit consistently.
- The team wanted to switch between AI assistants and agent frameworks without rebuilding the data layer each time.
The architecture
- One MCP server with typed tools for compute and container services, databases, caches, WAF rules and sampled requests, load balancers, log queries, metrics, security findings, and cost data.
- Adapters for the backend APIs of multiple platforms, normalised so an agent can compare the same figures across products without knowing each API.
- Authenticated end to end: every client session and every tool call is authorised against the operator's identity provider, with read-only defaults and an explicit confirm-and-reason step for any mutating operation.
- Client-agnostic by design: the same server serves Claude, ChatGPT-compatible clients, and custom agents, with every call audit-logged.
Model Context ProtocolTypeScriptOAuth 2.0AWSCloudWatchWAFClaudeOpenAI
The outcome
- Engineers and analysts ask questions across infrastructure, security, and product data in one place instead of across a dozen consoles.
- New AI tools plug into the estate in minutes rather than weeks, and the operator owns the integration layer rather than any single vendor.
Services drawn on
A similar problem on your roadmap?
A Discovery Sprint is the fastest path to a scoped, costed proposal.