How to write effective tool descriptions for Snowflake-managed MCP servers

📢 New Fabric Toolbox Release: DAX Performance Tuner MCP Server 🚀 With all of the buzz around MCP servers, I wanted to see if one could be created that would help you optimize DAX. 🛠️ Introducing: DAX Performance Tuner! The MCP server give LLMs the tools it needs to optimize your DAX queries using a systematic, research-driven process. How it works: After the LLM connects to your model, it prepares your query for optimization. This includes defining model measures and UDFs, executing the query several times under a trace, returning relevant optimization guidance, and defining the relevant parts of the model’s metadata. After analyzing the results, the LLM will attempt to optimize your query, ensuring it returns the same results. It is definitely not perfect, but I have seen some pretty impressive results so far. It helped optimize a 150-line query's duration by 94% (140s to 8s)! I would love to hear your feedback if you get a chance to test it out. 🙂 Link to repo ➡️ https://lnkd.in/eZZ2QQvN Link to demo video ➡️ https://lnkd.in/eArqsR2R

Last weekend, I shared a post about using AI to interact with Power BI models through MCP. While this isn't groundbreaking—there are plenty of similar solutions out there, and you can build your own—the possibilities are worth exploring. You can create an MCP with multiple tools or focus on a specific use case, like DAX optimization (see the attached post). What I particularly like about that approach, described in the post) is how it leverages a curated list of DAX optimization articles as a knowledge base for the AI. Test, experiment, and discover your own way of integrating AI into Power BI modeling. The potential is there—you just need to find what works for you.

"Our database is the bottleneck." Translation: "We never planned for success and now success is killing us." Before you panic and rebuild everything: - Are you asking your database stupid questions? (Fix the questions first) - Are you storing stuff you never actually use? (Delete it) - Are you checking the same thing 1000 times per second? (Cache it) 90% of database problems are solved by asking it to do less work, not buying bigger servers. What's the slowest part of your platform that customers actually notice?

The MCP Kool‑Aid is strong these days 😅 Most MCP servers are not only redundant—they might actually underperform existing options if your agent already has a built-in shell tool and a solid CLI or API is available for the task. Agents will load all MCP tool definitions at session start, whether they are used, burning context tokens for nothing. Even on a model with a theoretical 1M token window, your effective usable budget might be closer to 100k: throwing 10ks away on “just in case” tool definitions is a big deal. Agents can also pipe CLIs through jq, yq, grep, or custom filters, keeping the token footprint lean. MCP servers can however be marginally faster, as they sidestep e.g. Claude Code's system prompt to check for malicious commands. Remote MCP servers used to be easier to distribute/update compared to subagents, though Claude plugins have changed that. But ultimately, the real differentiator is tool design and token efficiency. Ironically, the least-discussed MCP feature might be the one that matters most: MCP Prompts. These let an MCP server act as a prompt templating engine, steering agents across multi-tool, multi-step workflows, making these real agents reasoning improvers rather than just a change of plumbing.

Bulletproof Backend — Day 4 / 100 Q. Can Database Replication really SPEED UP your app? A. Yes and No. What is it: Database replication means having identical copies of your database on multiple servers.  Why use it: 1. During high load, it optimizes read operations by distributing traffic across servers. 2. If one server goes down, others keep your app running — no downtime. When to use it: 1. Use replication when your database starts getting heavy read traffic. So yeah, if the load is really HIGH, yes. Otherwise not so much.

What happens when a new server joins a running system? Does all the data move around again? Or does only a small part shift to balance things out? Today I learned about a process that answers this perfectly -- Consistent Hashing. Here’s the idea in simple steps: 1)Servers are arranged on a circular hash ring. 2)Each piece of data finds a position on that ring. 3)Data is stored in the next server moving clockwise. 4)When a server joins or leaves, only a few data items change place. It’s simple but clever — even a small change, like adding one server, only shifts a tiny part of the data, yet the whole system stays balanced. #SystemDesign #BackendLearning #ConsistentHashing #LearningInPublic #BackendJourney

Postgres 17 vs 18. How does read performance compare between the two? What is the best server configuration to maximize performance? Check out our latest blog for detailed benchmarks. https://lnkd.in/gUCDFYmE

Zero to full-stack monitoring in under 5 minutes. 🚀 Netwarden makes infrastructure monitoring painless: Server Auto-discovery, auto-setup of monitors, dashboards and alerts. One 8mb open source agent in your host is all you need.

⏰ Ever needed to run tasks automatically on your server? Crontab makes it easy! Schedule commands and automate your server with ease. 🎥 Watch the full guide: https://ow.ly/No9v50Wi9ZH #Crontab #LinuxTips #ServerAutomation #Contabo #WebHosting

✅ Monitoring cron jobs is key to maintaining healthy server resources. Learn how to keep track of running cron on the Cloudways Platform 🔗 https://bit.ly/4hrfwBr