Enterprise Platform MCP
In developmentBuilt at work for a national healthcare data companyAI assistants reach a large healthcare data platform through four adapters: a command line, a local server, a hosted server behind company sign-in, and an Agent Skill. Thirteen tools sit behind three permission levels, and nothing reaches production until a person approves it.
What it does
- One shared core with four adapters: a command line, a local server, a hosted server behind company sign-in, and an Agent Skill.
- 13 tools under three permission levels, and nothing is promoted to production without a person approving it.
- A PHI guard in the core and an audit trail on every write.
From the app's own materials.
Who it is for
Engineering teams giving AI assistants access to a large healthcare data platform.
How it works
Ask
A person or an AI assistant asks for something through one of four adapters.
Core
The shared core applies the same policy and PHI guard to every request.
Tools
13 tools, under read, write, and promote permissions.
Approve
Promotion to production waits for a person, every time.
Audit
Every write is recorded in the audit trail.
The design
The decision that matters
One core for every adapter
The command line, the local server, the hosted server, and the Agent Skill all call the same core, so a different adapter never means a different set of rules.
Built with
- Model Context Protocol
- Python
- Agent Skills