Claude Delegator: Your AI Dream Team Inside Claude Code

Tired of hitting Claude's limits on complex tasks? What if you could summon a team of AI specialists—architects, security analysts, and code reviewers—directly inside your Claude Code workflow? That's exactly what Claude Delegator delivers. This revolutionary plugin transforms your solo Claude instance into a sophisticated multi-AI orchestration platform, automatically routing tasks to GPT or Gemini experts who analyze, implement, and synthesize solutions that single models simply can't match.

In this deep dive, you'll discover how Claude Delegator turns one AI into five domain masters, learn the exact installation commands, explore real code examples from the repository, and master advanced patterns that elite developers use to ship bulletproof code faster. Whether you're architecting distributed systems or hunting security vulnerabilities, this tool redefines what's possible in AI-assisted development.

What is Claude Delegator?

Claude Delegator is a native MCP (Model Context Protocol) plugin for Claude Code that creates a specialized team of AI subagents. Created by Jarrod Watts, this open-source tool bridges the gap between Claude's general-purpose capabilities and the deep expertise of purpose-built AI models. Instead of wrestling with a single AI that knows a little about everything, you get five dedicated experts—each armed with specialized system prompts—working in concert to solve your development challenges.

The plugin operates through the Model Context Protocol, a standardized way for AI assistants to interact with external tools and services. By registering Codex CLI (for GPT models) or Gemini CLI as MCP servers, Claude Delegator extends Claude Code's functionality without disrupting your existing workflow. When you ask a security question, it doesn't just prompt Claude harder—it delegates to a Security Analyst subagent powered by GPT-4 or Gemini, then synthesizes the expert's findings into actionable insights.

Why it's trending now: As developers push AI assistants beyond simple code completion into architecture decisions and security auditing, the limitations of single-model approaches become glaring. Claude Delegator solves this by embracing specialization over generalization. The repository has gained rapid traction because it demonstrates a practical implementation of AI agent orchestration—a hot topic in 2024's AI engineering landscape. It proves that the future isn't one super-AI, but coordinated teams of specialized agents, each excelling in their domain.

Key Features That Transform Your Workflow

Five Domain Experts at Your Command

The heart of Claude Delegator is its roster of specialists, each crafted with precision prompts in the prompts/ directory:

• Architect: Masters system design, tradeoff analysis, and complex debugging. When you're deciding between Redis and in-memory caching or designing a microservices topology, this expert provides battle-tested patterns and catches architectural anti-patterns before they become production nightmares.

• Plan Reviewer: Acts as your strategic advisor before you write a single line. Feed it a migration plan or implementation strategy, and it validates assumptions, identifies risks, and suggests optimizations. This preemptive review saves countless hours of rework.

• Scope Analyst: The ambiguity killer. When requirements feel fuzzy, this expert interrogates your specifications, surfaces hidden assumptions, and clarifies edge cases. It's like having a veteran product manager who never misses a detail.

• Code Reviewer: Your tireless pair programmer. It finds logical bugs, spots performance bottlenecks, enforces best practices, and suggests refactoring opportunities. Unlike linters, it understands intent and catches subtle errors that static analysis misses.

• Security Analyst: A white-hat hacker in your pocket. It performs threat modeling, vulnerability scanning, and security architecture reviews. From authentication flows to API endpoint hardening, it thinks like an attacker to protect your code.

Dual Provider Support: GPT or Gemini

Flexibility defines modern tooling. Claude Delegator supports both OpenAI's Codex CLI and Google's Gemini CLI, automatically detecting which provider you've configured. Use GPT-4 for its reasoning prowess, Gemini for its multimodal capabilities, or both for redundancy. The plugin routes tasks intelligently, falling back gracefully if one provider is unavailable.

Advisory vs. Implementation Modes

Every expert operates in two distinct modes:

• Advisory Mode (read-only sandbox): For analysis, recommendations, and reviews. The expert examines your code without making changes, delivering insights that preserve your agency as the developer.

• Implementation Mode (workspace-write sandbox): For hands-on fixes and code generation. The expert directly modifies files, implements solutions, and even retries failed attempts up to 3 times before escalating back to you.

Claude automatically selects the mode based on your request's intent. Ask "Is this secure?" and you get analysis. Say "Harden this endpoint" and watch the expert implement security headers, input validation, and rate limiting automatically.

Intelligent Auto-Routing and Synthesis

Raw AI output is often verbose and inconsistent. Claude Delegator's magic lies in synthesis, not passthrough. When an expert completes analysis, Claude interprets their findings, applies its own judgment, and presents a polished, actionable response. You get the best of both worlds: specialized expertise and Claude's coherent communication style.

The auto-routing engine detects task types through pattern matching. Mention "tradeoffs" or "structure" and the Architect activates. Say "review this PR" and the Code Reviewer engages. This happens seamlessly—no manual expert selection required.

Multi-Turn Conversation Chains

Complex tasks require context preservation. The plugin supports threaded conversations where experts remember previous turns. Start with architecture planning, move to implementation, then request security review—all within a single context chain. Each turn receives a threadId, enabling experts to build upon prior work without repeating context.

Real-World Use Cases Where Claude Delegator Shines

1. Architecture Decisions for Distributed Systems

You're building a notification service that must handle 10M daily pushes. Should you use WebSockets, Server-Sent Events, or polling? The Architect expert analyzes your throughput requirements, latency constraints, and infrastructure budget. It doesn't just list options—it models tradeoffs, calculates cost implications, and recommends a hybrid WebSocket approach with fallback polling. It even generates the initial service structure and deployment manifests.

Impact: Reduces week-long architecture debates to 15-minute expert consultations with production-ready scaffolding.

2. Pre-Implementation Plan Validation

Before migrating your monolithic API to microservices, you draft a detailed plan. The Plan Reviewer examines your service boundaries, data migration strategy, and rollback procedures. It spots that you forgot to account for distributed transactions in the checkout flow and that your proposed database-per-service pattern will break reporting queries. It suggests a saga pattern and read replicas before you cut any code.

Impact: Catches 80% of migration pitfalls in the planning phase, saving months of rework.

3. Security Auditing for Authentication Flows

You've implemented OAuth 2.0 with JWTs, but a nagging feeling persists: "Is this actually secure?" The Security Analyst performs threat modeling, identifies that your token refresh endpoint lacks rate limiting, discovers your JWTs aren't properly scoped, and notices you're using RSA512 when EdDSA would be more efficient. It then implements the fixes: adds rate limiting middleware, scopes the tokens, and updates the signing algorithm.

Impact: Transforms security from an afterthought into a proactive, automated process. Finds vulnerabilities that would cost $50K+ in a penetration test.

4. Code Review for Performance Bottlenecks

Your React component renders 500 rows of data and feels sluggish. The Code Reviewer profiles the virtual DOM diffing, spots that you're creating new function references on each render, identifies unnecessary re-renders from unstable prop references, and notices you're not using React.memo where appropriate. It refactors with useCallback hooks, stabilizes props with useMemo, and wraps components in memo, cutting render time by 70%.

Impact: Delivers senior engineer-level code review in minutes, not hours. Teaches optimization patterns while fixing issues.

5. Scope Clarification for Ambiguous Requirements

The product spec says: "Make the search faster." The Scope Analyst immediately flags the ambiguity: "Faster by what metric? Latency? Throughput? For what query volume? What's the current baseline?" It generates a list of 12 clarifying questions, suggests A/B testing criteria, and drafts a performance SLO document. Once you answer, it creates a detailed implementation ticket with acceptance criteria.

Impact: Eliminates scope creep and misaligned expectations. Ensures you're solving the right problem, not just a problem.

Step-by-Step Installation & Setup Guide

Getting started with Claude Delegator takes under five minutes. Follow these exact commands to activate your AI team.

Prerequisites

You need Claude Code installed and at least one of these providers:

• Codex CLI for GPT models: npm install -g @openai/codex
• Gemini CLI for Gemini models: npm install -g @google/gemini-cli

Authenticate each provider:

• For Codex: Run codex login and complete the OAuth flow
• For Gemini: Run gemini once to trigger sign-in, or set GOOGLE_API_KEY environment variable

Installation Steps

Step 1: Add the Marketplace

Inside any Claude Code instance, register the plugin marketplace:

/plugin marketplace add jarrodwatts/claude-delegator

This command tells Claude Code where to find the delegator plugin. It fetches metadata about available commands and MCP configurations.

Step 2: Install the Plugin

/plugin install claude-delegator

Claude downloads the plugin code to your local machine and registers its commands. You'll see a confirmation message listing available actions.

Step 3: Run Setup

/claude-delegator:setup

The setup wizard automatically:

• Detects installed providers (Codex and/or Gemini)
• Registers the appropriate MCP servers
• Configures default sandbox modes
• Validates authentication tokens
• Installs delegation rules that map user requests to experts

Done! Claude now routes complex tasks automatically. No restart required.

Manual MCP Setup (If Needed)

If the automated setup fails, manually register MCP servers:

# For Codex (GPT) - Idempotent, safe to rerun
claude mcp remove codex >/dev/null 2>&1 || true
claude mcp add --transport stdio --scope user codex -- codex -m gpt-5.3-codex mcp-server

# For Gemini - Idempotent, safe to rerun  
claude mcp remove gemini >/dev/null 2>&1 || true
claude mcp add --transport stdio --scope user gemini -- node ${CLAUDE_PLUGIN_ROOT}/server/gemini/index.js

Verify installation:

# List registered MCP servers
claude mcp list

# Test Gemini server health
printf '{