# Tuning Our MCP Server with Agent Benchmarks

DevFeed: [Tuning Our MCP Server with Agent Benchmarks](<https://devfeed.tech/articles/tuning-our-mcp-server-with-agent-benchmarks-116397.md>)

Original publisher: [Read original article](<https://upstash.com/blog/coding-agent-benchmark-mcp-server>)

Author: Cahid Arda Oz

Published: 2026-11-03T00:00:00Z

Content type: news

Language: en

Sources: [Upstash](<https://devfeed.tech/sources/upstash.md>)

Topics: [AI Agent](<https://devfeed.tech/topics/ai-agent.md>), [Benchmark](<https://devfeed.tech/topics/benchmark.md>), [Claude Code](<https://devfeed.tech/topics/claude-code.md>), [cursor](<https://devfeed.tech/topics/cursor.md>)

Tags: [agent](<https://devfeed.tech/tags/agent.md>), [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [benchmark](<https://devfeed.tech/tags/benchmark.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [coding](<https://devfeed.tech/tags/coding.md>), [coding-agents](<https://devfeed.tech/tags/coding-agents.md>), [fixed](<https://devfeed.tech/tags/fixed.md>), [mcp](<https://devfeed.tech/tags/mcp.md>), [mcp-server](<https://devfeed.tech/tags/mcp-server.md>), [redis](<https://devfeed.tech/tags/redis.md>), [results](<https://devfeed.tech/tags/results.md>), [tasks](<https://devfeed.tech/tags/tasks.md>)

## AI overview

A benchmark of 369 runs across seven coding agents found that most could complete Upstash Redis setup and rate-limiting tasks, but exposed four gaps: no daily-backup tool, confusion about Redis logs, delayed usage stats reported as zero, and deletes proceeding without confirmation. Upstash added a backup toggle, clarified log scope and an approximately two-minute stats delay, and made the first delete call non-destructive pending user confirmation. In reruns across five agent/model pairs, all answers to the weekly command-count question had been zero; afterward, four gave the right count and 11 reported that stats were still catching up. App setup and rate-limiting misses occurred only with Haiku 5.5; other tested models passed those tasks. Delete confirmation varied by agent, and the elicitation form required the person to type the database name.

## Source excerpt

We ran coding agents through real Upstash Redis tasks with the Upstash MCP server, fixed what the benchmark found, and compared the results before and after.