v0.87.0

Release Date: February 8, 2026

Note: this plugin is currently tested only on macOS.

Reporting issues

Use “Collect Logs and Diagnostics Data” action in IntelliJ to collect information about the problem. Review the log file for specific problem.

Discord Community

Message Eugene on LinkedIn for a Discord invite: linkedin.com/in/jonnyzzz

Direct invite link: discord.gg/e9qgQ7NeTC

That is the best place to discuss the plugin, ideas, features.

Download

• Download mcp-steroid-0.87.0.zip — Plugin ZIP
• Release page on GitHub

SHA-256: 7a7ee4edee94aa3a298c4543407e6e80723e9e26a2ca92fc3bdc09acb45abbc3

End User License Agreement (EULA) - By downloading and using this plugin, you agree to the terms of the EULA.

Installation

1. Download the .zip file above
2. In IntelliJ IDEA, go to Settings → Plugins → ⚙️ → Install Plugin from Disk…
3. Select the downloaded ZIP file
4. Restart the IDE

Connecting to Cloud Agents

See Getting Started.

Supported AI Agents

Claude Code, Codex, Gemini, Cursor, and any other coding agent that supports the latest MCP via HTTP Streaming.

What’s New in v0.87.0

Automatic Modal Dialog Closing (opt-in)

Modal dialogs (e.g. “Tip of the Day”, update prompts, trust dialogs) can block code execution silently. MCP Steroid can now automatically close modal dialogs before executing code via steroid_execute_code. Enable it with the mcp.steroid.dialog.killer.enabled registry key (disabled by default while the feature matures).

Skills & Resources System Overhaul

The entire skills and resources system has been rebuilt from the ground up:

• Build-time code generation — Prompt files in src/main/prompts/ are now compiled to Kotlin classes using KotlinPoet, providing type-safe, obfuscated access to all skill content
• Rich article model — Each resource has a header, payload, description, and cross-references (see-also links) that are generated automatically at build time
• Auto-generated table of contents — Each resource folder gets a TOC article, making it easy for agents to discover related content
• Skill catalog & prompt registry — Skills are registered as standard MCP prompts and resources, so agents discover them via the MCP protocol
• Improved prompt quality — Debugger, LSP, IDE, and VCS example resources have been updated for IntelliJ 253 APIs with fixed imports, resource leaks, and correct read/write action usage

Plugin No Longer Expires

The time-bomb mechanism from the v0.86.0 preview has been removed. The plugin no longer has a build expiration date.

Stability & Bug Fixes

• Fix for project paths with spaces — Kotlinc now uses an argfile for the classpath, correctly handling project paths that contain spaces
• MCP session stability — Server restart now properly recreates the Ktor server instance, fixing stale session issues
• MCP readiness check — Improved server readiness detection and compiler stderr handling
• Reduced storage noise — steroid_list_projects and steroid_list_windows calls are no longer logged to the .idea/mcp-steroid folder, keeping storage focused on actual executions

Improved Tool Output

• steroid_list_projects and steroid_list_windows now report IDE product name, build number, and PID
• Tool preference support — agents can express which tools they prefer to use

Docker-Based Integration Test Infrastructure

A comprehensive test framework for running AI agents inside Docker containers with a full IntelliJ IDE:

• IdeContainerSession — Manages the full lifecycle of an IntelliJ instance inside Docker with video recording
• Live video streaming — Watch agent sessions in real-time via an HTML dashboard
• Multi-agent support — Test with Claude Code, Codex, and Gemini CLIs in isolated containers
• IntelliJ Community Edition — Integration tests now run against Community Edition for broader compatibility

Website & Documentation

• Added Releases section to the website
• Published Strategy, Learning Methodology, and Experiments pages
• Improved docs layout with weight-based page ordering
• Video carousel on homepage

Feedback

We’re actively learning what content and features work best for the plugin. Your feedback is invaluable!

If you encounter any issues or have suggestions, please let us know in GitHub Issues! Your feedback directly shapes the development of MCP Steroid.