AI with clear boundaries
We connect company data to AI agents in a controlled way
MCP lets an agent use defined business tools without opening the database or granting unrestricted SQL access.
The rental platform uses a read-only MCP server with analytical tools and views that exclude PII.
Data sources
Read-only views without PII
MCP tools
Authorised agent
A controlled data layer
The agent requests a specific tool; the server validates parameters and returns only the intended result.
- + Purpose-built tools
- + Typed inputs
- + Limited outputs
- + Call logs
Metrics, not tables
We define revenue, cost, monthly and operational metrics so answers remain consistent.
- + One definition per metric
- + Authorised filters
- + Time context
Security by design
The analytics layer is read-only and its views exclude PII.
- + No writes
- + No arbitrary SQL
- + Authentication
- + Least privilege
AI clients
ChatGPT, Claude, Codex or Cursor can access tools only when MCP is supported and the connection is configured and authorised.
- + Explicit setup
- + Separate credentials
- + Revocable access
Frequently asked questions
Can the agent change data?
Not in this architecture. Analytical tools and views are read-only.
Does MCP expose the database publicly?
No. It exposes authenticated, controlled tools—not the database.
Can it run any SQL?
No. Queries have a defined scope and unrestricted SQL is deliberately unavailable.
Next step
JUYE / START
Chaos between systems, or manual work nobody wants to do? Write to us.
Tell us which systems you use and where work gets stuck.