claude-code-mode

Overview

claude-code-mode is a powerful plugin for Claude Code that optimizes the interaction with MCP (Multi-Cloud Protocol) tools by utilizing a single sandboxed JavaScript program. This innovative approach consolidates multiple tool calls into a streamlined process, minimizing the amount of context data that is passed around and enhancing the efficiency of tasks that require multiple calls.

Key Features

• Optimized Token Use: Instead of multiple full result contexts for each tool call, only the essential return value is sent back into the context, thereby reducing token usage and streamlining the model's focus on the task at hand.
• Sandboxed Environment: The program executes in its own Node.js process, ensuring a secure and controlled environment with limited resource access. On platforms like macOS and specific Linux distributions, it operates with further restrictions including no file system access and CPU limitations, enhancing security and performance.
• Smart Hinting: Each MCP server comes with succinct hints detailing result formats, required arguments, and constraints. The model can recommend new hints or eliminate incorrect ones, subject to user approval, ensuring accuracy and relevance.
• Reusable Functions: The model has the capability of proposing program segments as reusable functions for future sessions. Users can review and approve this code, promoting efficiency and adaptability in ongoing development tasks.
• User-Centric Permissions: Every MCP tool call made within the program adheres to the established permission rules, ensuring that security protocols are maintained throughout the process.

Example Usage

To illustrate, when posed with the question, "Which of my open Jira issues have no update in the last 14 days?", the model crafts a precise JavaScript program:

const issues = await call("mcp__claude_ai_Jira__searchJiraIssuesUsingJql", {
  cloudId: "example.atlassian.net",
  jql: "assignee = currentUser() AND statusCategory != Done",
})
const cutoff = Date.now() - 14 * 24 * 3600 * 1000
return issues.issues
  .filter(i => Date.parse(i.fields.updated) < cutoff)
  .map(i => ({ key: i.key, title: i.fields.summary }))

This method ensures that the context only receives the necessary brief list instead of the complete issue details, thereby enhancing performance.

Documentation

For detailed guidance, refer to the following documentation pages:

• Installation: Find requirements and installation instructions.
• Usage: Learn about tools, how to write programs, and handle results efficiently.
• Hints: Guidance on creating hints and safety rules.
• Saved Functions: Approve or create functions for recurring use.
• Permissions: Understand the principles of manual and automatic permission settings.
• Configuration: Detailed options and their defaults.
• How It Works: Insight into the process from start to finish.
• Troubleshooting: Solutions to common errors and limitations.
• Development: Information on layout, tests, and release management.