Master 5 New Features of Claude Code v2.1.92: MCP Result Persistence, Plugin Executables, and Interactive Tutorials

Author's Note: A detailed breakdown of the core updates in Claude Code versions v2.1.92 through v2.1.90, including MCP tool result persistence (500K characters), plugin bin directories, the /powerup interactive tutorial, and 5 other major feature upgrades.

Claude Code has released a series of significant updates over the last three versions (v2.1.90 ~ v2.1.92). Features like MCP tool result persistence, plugin executable directories, and the /powerup interactive tutorial have sparked widespread interest in the developer community. This article provides a quick look at how these updates impact your AI-assisted programming workflow.

Core Value: Get up to speed on the latest 3 versions of Claude Code in just 3 minutes, and master the MCP persistence configuration and the /powerup learning path.

Claude Code v2.1.92 Quick Look

Claude Code released three versions in quick succession in early April 2026: v2.1.90, v2.1.91, and v2.1.92. Here is a complete breakdown of the core updates across these versions:

Claude Code v2.1.92 Key Highlights

The core changes in v2.1.92 focus on enterprise-grade security controls and developer experience improvements:

The forceRemoteSettingsRefresh policy is a major update for enterprise teams. When enabled, the CLI forces a pull of the latest remotely managed settings every time it starts. If the pull fails, it exits (a fail-closed security model). This ensures that enterprise IT teams can guarantee all developers are consistently running configurations that comply with company security policies, preventing drift caused by local caching.

The Bedrock interactive configuration wizard significantly improves the first-time setup experience for AWS users. After selecting "3rd-party platform" on the login screen, the wizard guides you through the entire process of AWS authentication, region configuration, credential verification, and model binding, eliminating the need to manually edit configuration files.

Deep Dive into Claude Code MCP Tool Result Persistence

Core Mechanism of MCP Result Persistence

The MCP tool result persistence introduced in v2.1.91 is arguably the most impactful feature for developers in this release. Previously, large results returned by MCP tools—such as database schemas or extensive codebase indexes—would be truncated due to default limits, leading to a loss of critical context.

Now, MCP servers can override these default limits by setting the _meta["anthropic/maxResultSizeChars"] field, supporting up to 500,000 characters (approx. 500K). This means:

• Complete Database Schema Transfer: The full structure of large-scale databases can be passed to Claude in one go.
• No More Truncated Codebase Indexes: Project-level file indexes and dependency graphs can be fully preserved.
• Full API Documentation Loading: Comprehensive documentation for complex APIs can be ingested as context.

How to Configure MCP Result Persistence

Simply add the _meta field to your MCP server response to enable it:

{
  "content": [
    {
      "type": "text",
      "text": "... large result content ..."
    }
  ],
  "_meta": {
    "anthropic/maxResultSizeChars": 500000
  }
}

import { Server } from "@modelcontextprotocol/sdk/server/index.js";

const server = new Server({
  name: "database-schema-server",
  version: "1.0.0"
});

server.setRequestHandler("tools/call", async (request) => {
  if (request.params.name === "get_database_schema") {
    const schema = await fetchFullDatabaseSchema();

    return {
      content: [
        {
          type: "text",
          text: JSON.stringify(schema, null, 2)
        }
      ],
      _meta: {
        "anthropic/maxResultSizeChars": 500000
      }
    };
  }
});

🎯 Development Tip: If you're building an MCP server, be sure to set a reasonable maxResultSizeChars value. You don't always need to hit the 500K limit; just choose an upper bound that fits your actual data needs. For scenarios where you need to test across various AI models, you can use the APIYI (apiyi.com) platform to centrally manage your API keys and quickly verify how your MCP server performs with different models.

Claude Code Plugin Executable Support

Plugin bin/ Directory Mechanism

v2.1.91 adds support for a plugin bin/ directory, a major expansion for the Claude Code plugin ecosystem. Plugin developers can now include executable files in their packages and invoke them directly as bare commands within the Bash tool.

Typical use cases for this feature include:

• Custom CLI Tools: Plugin-bundled command-line tools that run directly in Claude's Bash environment.
• Build Scripts: Project-specific build, test, and deployment scripts.
• Data Processing Tools: Custom data transformation and formatting utilities.

Plugin bin/ Directory Structure

my-claude-plugin/
├── plugin.json
├── src/
│   └── index.ts
└── bin/
    ├── lint-check        # Custom linting tool
    ├── db-migrate        # Database migration script
    └── deploy-preview    # Preview deployment tool

Once configured, Claude Code can call these tools directly when executing Bash commands:

# Claude can call plugin executables directly in Bash
lint-check src/
db-migrate --status
deploy-preview --branch feature/new-api

Other Key Updates in Claude Code v2.1.91

💡 Pro Tip: The --resume fix addresses a widespread issue where session history chains would break if asynchronous writes failed silently during a resume. This has been completely resolved, allowing you to safely interrupt and resume long-running development sessions.

Claude Code /powerup Interactive Tutorial System

/powerup Feature Overview

Introduced in v2.1.90, /powerup is Claude Code's first built-in interactive tutorial system. It provides terminal-based lessons with animated demonstrations, covering 18 key topics ranging from beginner to advanced levels.

How to use /powerup

Simply type the following in your Claude Code terminal:

/powerup

The system will display an interactive menu where you can select specific lesson modules. Each lesson includes:

• Conceptual Explanation: The core principles and design philosophy behind the feature.
• Animated Demo: Real-time operation demonstrations right in your terminal.
• Hands-on Practice: Guided exercises to help you try things out in your own projects.

Recommended Learning Path for /powerup

We recommend the following learning paths based on your experience level:

Getting Started — Master these 3 fundamentals first:

1. How to write the CLAUDE.md project configuration file.
2. Using the /compact command to manage your context window.
3. Using Plan Mode to solve complex tasks step-by-step.

Leveling Up — Focus on automation capabilities:

1. Configuring Hooks to automate your workflow.
2. Using Sub-agents to handle tasks in parallel.
3. Extending Claude's tool capabilities with MCP servers.

Advanced Mastery — Unlock full potential:

1. Using Worktrees for parallel development in isolated environments.
2. Using /schedule to run remote agents on a timer.
3. Integrating SDK Headless Mode into your CI/CD pipelines.

🚀 Quick Start: If you need to test various Large Language Models while developing MCP servers, we recommend using the unified API from APIYI (apiyi.com). You can access mainstream models like Claude, GPT, and Gemini with just a single API key.

Claude Code v2.1.92 Performance Optimization Summary

Write Tool Performance Boost

v2.1.92 optimizes the diff calculation for the Write tool, resulting in a performance increase of approximately 60% when handling large files containing tabs, & symbols, and $ signs. This is a significant quality-of-life improvement for developers who frequently edit large configuration files, shell scripts, and templates.

v2.1.90 Performance Optimizations

Version v2.1.90 includes several underlying performance improvements:

Other Improvements in v2.1.92

• /cost command enhancement: Subscribers can now view cost breakdowns by model and cache hit rate.
• /release-notes upgrade: Now an interactive version selector, making it easier to view update logs for any version.
• Remote control session naming: Defaults to using a hostname prefix (e.g., myhost-graceful-unicorn), with support for customization via --remote-control-session-name-prefix.
• Cache expiration notice: Pro users will see a footer prompt after the prompt cache expires, showing the approximate number of uncached tokens.

💰 Cost Tip: The Claude Code /cost command now provides cost analysis broken down by model. If you're using multiple AI models for development, you can use the billing features on the APIYI (apiyi.com) platform to centrally monitor usage and expenses for all models, making cost management much easier.

description: "A breakdown of the latest bug fixes and improvements in Claude Code v2.1.92, along with answers to common user questions."

Claude Code v2.1.92 Bug Fixes

We've addressed several issues over the last three releases that were impacting the user experience:

v2.1.92 Fixes

• Fixed an issue where sub-agent tmux sessions failed to create.
• Fixed an API 400 error caused by thinking blocks containing only whitespace.
• Fixed accidental survey submissions triggered by keypresses in autonomous mode.
• Fixed duplicate message displays during full-screen scroll playback.
• Fixed the plugin MCP server getting stuck in a "connecting" state.
• Removed the /tag and /vim commands.
• Restored Unix Socket blocking in the Linux sandbox.

v2.1.91 Fixes

• Fixed session history loss caused by --resume asynchronous write failures.
• Fixed cmd+delete shortcut issues in iTerm2, kitty, WezTerm, and Ghostty terminals.
• Fixed the loss of plan mode state after container restarts.

v2.1.90 Fixes

• Fixed a regression where --resume caused a full prompt-cache miss (introduced in v2.1.69).
• Fixed infinite rate-limit dialog loops.
• Fixed issues where autonomous mode ignored explicit user boundaries (e.g., "don't push").
• Fixed "File content has changed" errors from Edit/Write tools when the format-on-save hook is enabled.

FAQ

npm install -g @anthropic-ai/claude-code@latest

Summary

Here are the core updates for Claude Code versions v2.1.90 through v2.1.92:

1. MCP Result Persistence: Now supports tool results up to 500K characters, fixing issues where large schemas and indexes were previously truncated.
2. Plugin bin/ Directory: Plugins can now include executable files, significantly expanding the toolchain capabilities of Claude Code.
3. /powerup Interactive Tutorials: An 18-lesson course across three difficulty levels to help users of all skill levels master Claude Code systematically.
4. Enterprise Security Enhancements: The forceRemoteSettingsRefresh policy ensures configuration consistency across environments.
5. Comprehensive Performance Optimization: The Write tool is now 60% faster, and SSE transmission has been optimized from O(n²) to O(n).

These updates signal that Claude Code is evolving from a simple AI programming assistant into a full-fledged development platform. The continuous enhancement of the MCP ecosystem and the refinement of the plugin system provide developers with greater flexibility to build custom workflows.

We recommend using APIYI (apiyi.com) to quickly experience the API interfaces for Claude and other mainstream Large Language Models. The platform offers free credits and a unified interface for multiple models, making it easy to verify and compare performance.

📚 References

1. Claude Code Official Changelog: Complete version history.

   • Link: code.claude.com/docs/en/changelog
   • Description: The primary source for the latest version update details.

2. Claude Code GitHub Repository: Open-source code and issue tracking.

   • Link: github.com/anthropics/claude-code
   • Description: View source code, submit bug reports, and suggest new features.

3. MCP Protocol Specification: Official documentation for the Model Context Protocol.

   • Link: modelcontextprotocol.io
   • Description: The authoritative reference for MCP server development and tool result formatting.

4. Claude Code Plugin Development Guide: Getting started with plugin development.

   • Link: code.claude.com/docs/en/plugins
   • Description: Learn about plugin structure, bin directory configuration, and the publishing process.

Author: APIYI Technical Team
Technical Discussion: Feel free to share your experiences with Claude Code in the comments. For more AI development resources, visit the APIYI documentation center at docs.apiyi.com.