Triage AWS costs with a slimmer MCP server
The same AWS MCP server, built twice (bloated, 15 tools, and lean, 4 tools), so you can watch your AI agent get cheaper and smarter, live, in Kiro's /context view. Task: The same AWS MCP server, built twice (bloated, 15 tools, and lean, 4 tools), so you can watch your AI agent get cheaper and smarter, live, in Kiro's /context view. ⭐ If this helped you design better MCP tools, give it a star! It helps others find it. Video Tutorial • The Problem • The Scoreboard • Getting Started • The 4 Fixes • Contributing 🎬 Video Tutorial Watch the full rebuild, from a bloated 15-tool server that makes the agent flail, to a lean 4-tool server that answers in one clean call: 🎥 Watch on YouTube (link goes live with the episode) In the video you'll see: Why a typical MCP server quietly eats your context window before you ask a single question The two failure modes AWS named: bloat and confusion The bloated server making the agent take 7 tool calls to answer one question The…; tools: custom-built or unnamed agent. Human role: The original post does not say how a person stayed involved.; limitations: The demo used a small number of tools; the savings scale with larger production setups (90+ tools). The lean server initially lacked explicit handling for partial failures across regions (e.g., throttling), which could lead to incomplete audits being reported as complete. The author acknowledged this gap and planned to add scan status indicators in future versions..
Tools
custom-built or unnamed agent
Published on Agent Work Atlas:
Overview
The same AWS MCP server, built twice (bloated, 15 tools, and lean, 4 tools), so you can watch your AI agent get cheaper and smarter, live, in Kiro's /context view.
Workflow
The same AWS MCP server, built twice (bloated, 15 tools, and lean, 4 tools), so you can watch your AI agent get cheaper and smarter, live, in Kiro's /context view. ⭐ If this helped you design better MCP tools, give it a star! It helps others find it. Video Tutorial • The Problem • The Scoreboard • Getting Started • The 4 Fixes • Contributing 🎬 Video Tutorial Watch the full rebuild, from a bloated 15-tool server that makes the agent flail, to a lean 4-tool server that answers in one clean call: 🎥 Watch on YouTube (link goes live with the episode) In the video you'll see: Why a typical MCP server quietly eats your context window before you ask a single question The two failure modes AWS named: bloat and confusion The bloated server making the agent take 7 tool calls to answer one question The…
Reported result
The lean server reduced tool calls from 7 to 2 for the same query and shrank tool schemas by 88%. Both versions identified the same wasted resources (e.g., unattached volumes, orphaned snapshots). The author noted that while schema size decreased significantly, the overall context window usage in Kiro only dropped from 1.1% to 0.8% in this small demo, as other factors dominated the budget.
Human role
The original post does not say how a person stayed involved.
Limitations
The demo used a small number of tools; the savings scale with larger production setups (90+ tools). The lean server initially lacked explicit handling for partial failures across regions (e.g., throttling), which could lead to incomplete audits being reported as complete. The author acknowledged this gap and planned to add scan status indicators in future versions.
Try it
I want to replicate the AWS MCP server optimization experiment. Please clone the repository at https://github.com/simplynadaf/aws-lean-mcp.git. First, run the smoke test script to verify the tool count difference (15 vs 4) and schema size reduction. Then, configure both server versions in my Kiro environment. Use the prompt 'where am I wasting money in this account?' against a read-only AWS account in us-east-1. Record the number of tool calls and the final findings for both versions. Ensure the agent does not perform any write operations. If the lean server returns incomplete results due to API throttling, flag this as a known limitation. Do not publish any changes to the AWS account.
Source reports are not independently verified. Publication date is the catalog publication date, not the original article date.