The most useful MCP servers are not the flashiest. They are the ones that slot into an existing developer workflow and remove a manual step. io.github.duriantaco/skylos is a dead code detector that an AI agent can call before refactoring — find what is unused, then clean it up. Score 73.
What Skylos Does
Skylos uses static analysis to find unused functions, classes, methods, imports, and variables in Python codebases. Unlike coverage-based approaches that require a test suite to run and only detect code not exercised by tests, Skylos analyzes the code statically — it parses the Abstract Syntax Tree and traces references across the entire codebase. If a function is defined but never called anywhere, Skylos finds it. If a class is imported but never instantiated, Skylos flags it.
The name comes from the Greek word for "dog" — like a hunting dog sniffing out dead code. The project has 332 stars, 10 forks, and a release cadence of 15 per year, indicating active and responsive development. Two contributors. Apache-2.0 license.
The MCP Integration
As an MCP tool, the workflow becomes straightforward: before an AI agent refactors a Python module, it calls Skylos to identify what is already dead. The agent gets a structured report of unused code — functions nobody calls, imports nothing references, classes that exist but serve no purpose. Then it can clean up with confidence, knowing it is removing code that is provably unreachable rather than guessing based on context window analysis.
This addresses a real problem. AI coding assistants frequently hesitate to remove code because they cannot be certain it is unused — the function might be called from a file outside their context window, or imported dynamically, or referenced in a test they have not seen. Skylos provides the definitive answer by analyzing the full codebase, not just what the LLM can see.
The Numbers
| Metric | Value |
|---|---|
| Trust score | 73 |
| Stars | 332 |
| Forks | 10 |
| Watchers | 4 |
| License | Apache-2.0 |
| Active commit weeks | 34 |
| Contributors | 2 |
| Releases/year | 15 |
| Secrets | 1 |
| Flags | None |
The 34 active commit weeks out of 52 is strong for a two-person project — this is not abandonware. One secret is required, likely a configuration key. No flags.
Skylos represents the category of MCP servers that will quietly drive protocol adoption: small, focused developer tools that do one thing well and integrate cleanly into AI-assisted coding workflows. Not every MCP server needs to bridge a quantum computer or an enterprise platform. Sometimes the highest-value integration is the one that saves an agent thirty seconds of uncertainty about whether a function is still in use.
Score: 73. No flags. Apache-2.0.
Sources: duriantaco — GitHub · Scorecard: io.github.duriantaco (score 73)