The Unix Philosophy for AI Tools
The Unix philosophy of "do one thing well" turns out to apply perfectly to MCP servers. A server with three tightly-focused tools (query database, describe schema, explain query plan) is almost always more reliable than one with twenty tools covering database access, data visualization, report generation, email delivery, and file export.
The reason is straightforward: focused servers have fewer things that can go wrong. Each tool has been tested thoroughly because there are only three of them. The maintainer understands the full surface area. Dependencies are minimal because the functionality is narrow.
Model Selection Accuracy
When an AI model chooses which tool to use, it reads tool descriptions and matches them against your request. Three clear, distinct tools are much easier to match than twenty overlapping tools with subtle differences. The model makes better selections, which means you get better results.
A server with tools named "query_database," "describe_table," and "explain_query" is unambiguous. A server with tools named "get_data," "fetch_info," "retrieve_records," and "pull_results" creates confusion about which one to use when.
Maintenance Reality
Small servers are easier to maintain. The maintainer can hold the entire codebase in their head. Dependencies stay minimal. Tests cover the full surface area. When something breaks, the fix is usually quick because the codebase is small enough to understand completely.
Large, ambitious servers often suffer from partial abandonment. The core features work, but edge features stop getting maintained. Half the tools work great; the other half haven't been updated in months. The overall security grade drops because the maintainer can't keep up with all the dependencies.
Composability
Small servers compose beautifully. Connect a focused database server, a focused file server, and a focused search server, and the AI model orchestrates them together for complex tasks. You get the flexibility of many capabilities without any single server trying to be everything.
This composability is one of MCP's greatest strengths. The protocol was designed for many small servers working together, not for monolithic servers doing everything. Working with the protocol's design rather than against it produces better outcomes.
Related Reading
- The Beginner's Mistake of Over-Connecting MCP Servers
- How AI Assistants Choose Which Tool to Use
- The Real Difference Between MCP Resources and Tools
Browse MCP servers on Skillful.sh. Search 137,000+ AI tools.