“Should we use MCP or a REST API?” is usually the wrong first question.
The better question is:
What boundary are you trying to give the agent?
REST works well when the contract is stable, resource-oriented, and shared by normal software clients: dashboards, partners, mobile apps, backend services.
MCP is different. It is useful when the client is an AI agent that needs:
- discoverable tools
- tool descriptions
- structured parameters
- permission context
- policy-aware errors
- audit-friendly tool calls
- task-oriented workflows
For database access, that difference matters.
An agent should not just know “there is an endpoint.” It should know which tool is safe for the task, what scope applies, what result shape to expect, and what to do when the request is too broad.
Often the clean answer is both:
REST behind the service boundary.
MCP at the agent boundary.
Longer version: MCP vs REST API for AI agents
Do not choose the fashionable protocol. Choose the boundary you want the agent to respect.
Top comments (0)