claude-code-bridge

Claude Code Bridge

Delegate coding tasks to Claude Code CLI from any context.

When to Use

• Parallel Work: Spawn Claude Code to work on a subtask while you continue
• Different Context: Need Claude Code's file access in a specific directory
• Specialized Tasks: Let Claude Code handle complex refactoring, testing, or debugging
• Isolation: Keep risky operations in a separate session

Usage Patterns

Via MCP Server (Recommended)

If the claude-code-mcp server is configured, use the MCP tools:

Use claude_prompt to ask Claude Code: "Refactor the authentication module to use JWT"

Tools available:

• claude_prompt - Send a prompt to Claude Code
• claude_continue - Continue most recent conversation
• claude_resume - Resume a specific session by ID
• claude_version - Get version info

Via CLI (Direct)

Execute Claude Code directly:

# Single prompt
claude -p "List all TODO comments in this project" --output-format json

# Continue conversation
claude -p "Now fix the first TODO" --continue

# Resume specific session
claude -p "What files did we modify?" --resume <session-id>

Via Rust Library

use claude_code_rs::ClaudeCodeClient;

let client = ClaudeCodeClient::new()?;
let response = client.prompt("Analyze this codebase").await?;
println!("{}", response.text());

Configuration

MCP Server Setup

Add to Claude Desktop's claude_desktop_config.json:

{
  "mcpServers": {
    "claude-code": {
      "command": "npx",
      "args": ["@svnbjrn/claude-code-mcp"]
    }
  }
}

Or for Claude Code's ~/.claude/settings.json:

{
  "mcpServers": {
    "claude-code-bridge": {
      "command": "node",
      "args": ["/path/to/mcp-server/dist/index.js"]
    }
  }
}

Best Practices

1. Specify Working Directory: Always set working_dir to give Claude Code proper context
2. Use Descriptive Prompts: Be specific about what you want done
3. Capture Session IDs: Save session IDs from responses to resume conversations
4. Set Timeouts: Complex tasks may need longer timeouts (default: 5 min)
5. Check Results: Verify Claude Code's output before proceeding

Example Workflows

Code Review Delegation

I'll use Claude Code to review the PR changes:

[claude_prompt]
prompt: "Review the diff in pr-123.patch for security issues and code quality. Provide specific feedback."
working_dir: "/home/user/project"

Test Generation

Let me have Claude Code generate tests:

[claude_prompt]  
prompt: "Generate comprehensive unit tests for src/auth/*.ts. Use vitest and aim for 90% coverage."
working_dir: "/home/user/project"

Refactoring with Followup

[claude_prompt]
prompt: "Identify all uses of deprecated API v1 in this codebase"
working_dir: "/home/user/project"

# After reviewing results:
[claude_continue]
prompt: "Now migrate the first 3 files you found to API v2"

Error Handling

• Not Found: Ensure Claude Code is installed (npm install -g @anthropic-ai/claude-code)
• Timeout: Increase timeout_secs for complex tasks
• Permission Denied: Claude Code may need approval for file writes; use interactive mode for sensitive operations