Project Assessment — 2026-02-22

Project Identity

Development Velocity & Effort

Development pattern: solo developer operating at extremely high velocity with sustained daily commitment. The 25 commits/day average with atomic, descriptive messages indicates a disciplined, professional workflow.

Codebase Size & Structure

Modules (8 Gradle subprojects + buildSrc)

Code Metrics

Architecture Assessment

Core Design: 9/10

The architecture follows a layered pipeline with clean separation:

Agent (Claude/Codex/Gemini)
  | MCP JSON-RPC over HTTP/SSE
SteroidsMcpServer (Ktor)
  | Tool dispatch
ExecuteCodeToolHandler -> ExecutionManager
  | Two-phase execution
CodeEvalManager (external kotlinc compile) -> ScriptExecutor (coroutine run)
  | Results
ToolCallResult -> Agent

Novel Technical Decisions

Two-phase compile-then-execute. Separates Kotlin compilation (external process with daemon recovery) from execution (coroutine-based with timeout). Agents get compilation errors before runtime begins.

Modal dialog race detection. Uses Kotlin select{} to race script execution against IDE dialog appearance. If a dialog pops up, execution cancels and a screenshot is returned to the agent.

External Kotlin compiler isolation. Runs kotlinc as a separate process so agent scripts cannot starve the IDE’s own Kotlin daemon. Includes automatic daemon recovery on “Service is dying” errors.

Transport-agnostic MCP core. McpServerCore is pure JSON-RPC with zero HTTP dependencies. Transport (currently Ktor HTTP/SSE) is pluggable for future stdio or gRPC transports.

Human review workflow. Gate between agent code and execution with diff generation, timeout-based approval, and configurable trust levels.

Append-only execution storage. Immutable audit trail of every script, compilation output, and execution result under .idea/mcp-steroid/.

Build-time prompt compilation. Markdown prompt files compiled to obfuscated Kotlin classes with content validation tests auto-generated.

Code Quality Patterns

Coroutine discipline:

• Zero runBlocking() in production code
• SupervisorJob for storage I/O isolation
• withContext(Dispatchers.EDT) for UI access
• select{} for races (modal dialog detection)

IntelliJ Platform best practices:

• Proper readAction{} / writeAction{} for threading
• service<T>() for dependency lookup (no singletons)
• @Service(Service.Level.PROJECT) annotations
• Disposable lifecycle for cleanup
• ProcessCanceledException always rethrown

Error handling:

• Multi-layer with smart recovery (Kotlin daemon, incomplete code detection)
• Fast-fail semantics (compilation errors reported before execution)
• Modal dialog detection with screenshot capture
• SupervisorJob ensures storage writes complete even on cancellation

Testing Sophistication

Test Strategy: 8.5/10

The Docker-based AI agent integration tests launch a full IntelliJ IDE in a Docker container with VNC display, connect real AI agents, and verify end-to-end MCP workflows. The arena tests run curated project benchmarks comparing agent performance with and without MCP Steroid.

Commit Quality & Themes

Commit Discipline

• Atomic commits with descriptive messages
• Conventional-commit-like prefixes (feat(), fix(), refactor(), test(), docs())
• Task tracking references (T1-T7, R3-R14)
• Linear history with no merge commits

Development Themes

Quality Scorecard

Overall: 8.8/10 — Production-grade, well-architected MCP server plugin with several novel approaches to IDE integration and error resilience.

Competitive Positioning & Value

What Makes This Unique

MCP Steroid occupies a category-defining position: it is the bridge between AI coding agents and professional IDE capabilities. While agents typically work at the file-editing level (text in, text out), MCP Steroid gives them access to:

• Code intelligence — find usages, go to definition, symbol search
• Refactoring — rename, extract method, move
• Compilation & build — real Kotlin/Java compilation with error reporting
• Debugging — breakpoints, step-over, variable inspection
• Code analysis — inspections, warnings, quick-fixes
• Visual IDE — screenshots, component trees, OCR, input simulation
• Human oversight — code review before execution

Estimated Development Value

Executive Vision

MCP Steroid transforms AI coding agents from text editors into IDE-native developers.

Today’s AI agents (Claude, Codex, Gemini) work by reading and writing files. They cannot compile, debug, refactor, or run code analysis — the core activities of professional software development. MCP Steroid bridges this gap by exposing the full IntelliJ IDEA runtime to agents via the Model Context Protocol.

Key differentiators:

• Only product providing visual IDE access (screenshots + input dispatch) to AI agents
• Transport-agnostic MCP server running inside the IDE process with full API access
• Multi-agent support — works with Claude Code, OpenAI Codex, Google Gemini CLI
• Human-in-the-loop review — configurable approval gates before code execution
• Battle-tested — Docker-based integration tests with real agents on real projects

Market position: First-mover in the “AI-IDE bridge” category. As AI agents become the primary interface for software development, the IDE becomes the execution environment they need but cannot access. MCP Steroid is that access layer.

Current state: v0.88.0, actively developed at high velocity, comprehensive test coverage, production-ready for IntelliJ IDEA 2025.3+, with expansion to GoLand and WebStorm underway.

Growth vectors: Multi-IDE support (already started), NPX proxy for multi-instance aggregation, arena benchmarking framework for agent quality measurement, enterprise deployment via custom plugin repository.

Support the Project

MCP Steroid is built and maintained by a solo developer. Continued development, testing infrastructure (Docker-based agent tests are compute-intensive), and multi-IDE expansion all require sustained funding.

How you can help:

• Sponsor on GitHub — github.com/sponsors/jonnyzzz
• Submit real-world scenarios — share your repositories and workflows so we can benchmark and improve. See Support the Project for the submission template
• Share usage logs — the .idea/mcp-steroid folder in your project contains tool call logs that help us fine-tune prompts and skills
• Report issues — github.com/jonnyzzz/mcp-steroid/issues
• Join the community — Discord server or message Eugene on LinkedIn

Engineering leaders and sponsors interested in pilot evaluations or benchmark expansion are welcome to reach out directly.