Shrink an AWS cost-triage MCP server

They rebuilt a read-only AWS cost checker with far fewer tools and cut the calls needed to answer a waste question from seven to two.

Done with Custom agent· Intermediate· Oct 2026

What they did
They made the same MCP server two ways against one live AWS account: first as 15 small boto3 wrappers, then as 4 broader tools focused on outcomes. In Kiro, both versions answered the same question about where money was being wasted. The smaller design used a tool that handled the internal checks and returned five fields for each finding, plus tighter inputs and a separate savings estimate tool.
How it went
The original setup took 7 tool calls, while the slimmer version finished in 2. The tool descriptions dropped from 7,898 characters to 897, an 88% reduction, and both versions identified the same three wasteful resources. The Elastic IP charge was still not confirmed.
Tips
A smaller tool set helps most when you have lots of tools, not just on a tiny test. Keep the server read-only and compare both call count and schema size before and after.

Try it yourself with Claude Code

Audit my MCP server in [path] that exposes [number] tools for [AWS cost data or similar]. First, map which tools overlap, which ones can be merged, and how to reduce the schema size without losing the ability to answer [example question]. Stop and show me the proposed tool list, the expected before/after call count, and any tradeoffs before changing files. If I approve, refactor the server, then run the same question and report the tool calls and findings, but do not send, delete, or modify anything without asking me first.

Fill in the [brackets]. Agents can make mistakes, so review anything before it is sent, paid or deleted.

Read the original on dev.to ↗

Get 3 agent uses like this every morning

With the day's top AI news. Free, and one click to unsubscribe.