What MCP does, when a direct connection is simpler, and how to give Strawberry useful tool access without losing control.
What MCP adds
The Model Context Protocol gives an AI product a standard way to discover and call tools exposed by another system. An MCP server describes the tools it offers. Strawberry acts as the client that can use those tools during a task.
MCP is most valuable when you need a custom system, a local developer tool, or a service that does not already have a suitable native connection. For common apps, Strawberry’s built-in connectors are usually the shorter route.
Choose the smallest useful connection
Start from the result, not from the desire to connect everything. List the few reads and writes the workflow needs, decide which actions require approval, and expose only those capabilities.
Keep credentials out of prompts and source files. Use the connection’s supported authentication mechanism, review the server before trusting it, and test with non-sensitive data before using customer information.
- Prefer read-only tools for the first test.
- Use narrow names and descriptions so the right tool is easy to choose.
- Return structured errors instead of empty success responses.
- Require explicit review for messages, deletions, transactions, and other consequential writes.
Test the real workflow
A connection is not finished when the tools appear in a list. Run one representative task from start to finish. Check identity and account selection, permission failures, missing data, rate limits, and what the user sees before an external change.
Once the connection behaves reliably, pair it with the Official Skill or custom skill that explains the actual workflow. The tool supplies access. The skill preserves how your team wants the work done.