Problem
Every assistant integrates the same sources its own way. Permissions, logs and costs are scattered and nobody knows what goes into each prompt.
Solution
Each source becomes an MCP server. A gateway in front applies authentication, authorization, masking and logging before any context reaches the model.
When to use
- →More than one assistant or agent uses the same sources
- →There is sensitive data or audit requirements
- →You want to switch models without rewriting integrations
## Trade-offs
- +Reusable integrations
- +Governance and auditing in one place
- +Swap models without rewrites
- −A new layer to operate and observe
- −A central point that needs high availability