Thanks for support..
Release: "Sentinel-IR-MCP v0.2.2 on GitLab" ([https://gitlab.com/jakubrojicek11/sentinel-ir-mcp/-/tree/v0.2.2?ref_type=tags](https://gitlab.com/jakubrojicek11/sentinel-ir-mcp/-/tree/v0.2.2?ref_type=tags))
Package: "[sentinel-ir-mcp@0.2.2](mailto:sentinel-ir-mcp@0.2.2) on npm" ([https://www.npmjs.com/package/sentinel-ir-mcp/v/0.2.2](https://www.npmjs.com/package/sentinel-ir-mcp/v/0.2.2))
I'm building Sentinel-IR-MCP, a tool that helps developers connect an AI coding assistant to a code-analysis server without wrestling with manual configuration.
Here's the problem: making an MCP server work is one thing. Making installation reliable across operating systems, shells, paths with spaces or special characters, and different AI clients is another.
On macOS, I've tested Node.js 20, 22, and 24. On Windows, my latest compatibility run reported 45/45 successful command-quoting checks across Windows PowerShell 5.1, PowerShell 7, and "cmd.exe".
But testing commands isn't the same as proving every desktop client works perfectly. There are still gaps to investigate.
So, where should I focus next?
And a second question: would you actually use a tool that makes connecting AI coding assistants to code-analysis tools easier, or is MCP setup already simple enough for you?
I'm looking for real developer feedback before deciding where to spend the next round of testing.
...
I'm Testing MCP on Windows and macOS. Should I Support Both?
The problem isn't just starting the server
I'm building Sentinel-IR-MCP, a small project designed to make code analysis available to AI coding assistants through the Model Context Protocol (MCP).
The idea is straightforward: instead of making an AI assistant rediscover a codebase from raw files every time, Sentinel-IR can provide structured analysis through tools such as "analyze_code", "analyze_file", and "analyze_directory".
But there's another problem worth solving.
Connecting an MCP server should be easy. In practice, the setup can involve configuration files, shell quoting, command-line differences, application-specific settings, and paths containing characters that developers use every day.
A path like this is perfectly legitimate:
"C:\sentinel-test\it's $HOME & 100% ^caret! (x)" Yet a generated command can break if the wrong quoting rules are used. And a command that works in PowerShell may behave differently in "cmd.exe" or another environment.
For a developer, this is an annoying setup problem. For someone who wants to use AI coding tools without becoming an expert in their configuration, it can become a barrier. What have I tested so far?
On macOS, I ran the automated test suite on Node.js 20, 22, and 24. Each run passed 56 tests. I also tested configuration and connection behavior across several AI clients. Some paths worked, while others still need investigation.
On Windows, the latest run used Windows Server 2022, Node.js 22, Windows PowerShell 5.1, PowerShell 7, and "cmd.exe".
The command-quoting probe reported 45 successful checks and zero failures across those three shells, including paths with spaces, apostrophes, "%", "&", "^", and other special characters.
The compatibility smoke test also reported 12 passes and zero failures. It produced six informational results, though, so a successful exit code does not mean every scenario is fully verified.
I also verified a native Claude Code installation without signing in and confirmed that the configured server was reported as connected for one Windows path.
These results are evidence of progress, not proof that every client, shell, and operating-system combination works. GUI clients and actual tool calls still need separate testing.
There is one more important detail: the local code under test was version "0.2.3", while the published "sentinel-ir-mcp@0.2" package resolved to version "0.2.2" during the compatibility run. That distinction matters when claiming that a released version has been tested.
The decision I'm trying to make
I could focus on one operating system first, finish the complete installation and verification experience, and expand later.
Or I could treat Windows and macOS as first-class targets from the beginning, which means more compatibility testing and more edge cases to maintain.
I'm not asking which operating system is technically better. I'm trying to understand where a tool like this would actually be useful to developers.
Three questions for the community
Would you use a tool that helps configure and verify an MCP code-analysis server for your AI coding assistant, or do you already find this workflow simple enough?
Would you expect it to configure the client automatically, verify that the server starts and exposes its tools, test an actual analysis call, preserve existing configuration, or all of the above?
I'm building this in small steps and testing behavior rather than assuming that a successful installation means everything works.
Your answers will help me decide whether to polish one platform first or invest in cross-platform support now.