
Claude Delegator
https://github.com/jarrodwatts/claude-delegator- Category
- AI Agents
- Rank
- No. 1029Tools index
- Pricing
- Open Source
- Type
- TOOL
- Builder
- jarrodwatts
- GitHub
- 996 stars
- Date
About
MCP plugin that lets Claude Code delegate tasks to Codex and Gemini without leaving the session.
What it does
Claude Delegator turns architecture, planning, scope, code quality, and security requests into specialist consultations. Claude selects a role, sends structured context to Codex or Gemini, then interprets the result. Specialists can advise with read-only access or modify the workspace. Follow-up turns retain context through a returned thread identifier.
Why it's ranked here
The design has useful boundaries, not just model switching. Five defined roles, explicit delegation triggers, separate advisory and implementation modes, and synthesized responses make the workflow understandable. Multi-turn retries preserve context. The main reservation is operational weight: users still need Claude Code, at least one external provider CLI, authentication, MCP registration, and installed orchestration rules.
What's good
Provider choice is practical: Codex uses its native MCP server, while Gemini gets a zero-dependency bridge. The bridge validates required inputs, rejects unsupported sandbox values, avoids shell execution, and returns structured errors. Prompts tell specialists what to prioritize and when not to intervene. The rules also discourage delegation for trivial work, research, and first attempts at fixes.
Tradeoffs
Setup changes user-scoped MCP configuration and copies orchestration rules into the Claude configuration area. Provider CLIs and their authentication remain separate prerequisites. Automatic routing depends on semantic trigger rules, so users may need an explicit request when an expert does not activate. Implementation mode grants workspace write access. Gemini supports only read-only and workspace-write sandbox choices, and its bridge parses JSON from CLI output.
How to use it well
Use it for consequential design choices, plan validation, ambiguous requirements, security checks, code review, and debugging after repeated failures. Start advisory work read-only, reserve write access for requested fixes, and keep related implementation steps in one thread. It suits Claude Code users who already want Codex, Gemini, or both installed. It does not replace research tools, routine file operations, or a first debugging attempt.
Technical notes+
README.md documents marketplace installation, setup, dual providers, specialist routing, sandbox modes, and multi-turn sessions. commands/setup.md registers user-scoped stdio MCP servers and installs Markdown rules; commands/uninstall.md removes both. rules/orchestration.md defines routing, prompt injection, synthesis, retry limits, and threadId continuation. rules/triggers.md supplies semantic and explicit activation rules. server/gemini/index.js is a Node.js JSON-RPC 2.0 stdio bridge that spawns Gemini without a shell, exposes new-session and reply tools, validates arguments, and uses only built-in Node modules. config/providers.json records provider commands and specialist coverage. CLAUDE.md states there is no build step or dependency installation for the plugin itself.
Observed
- License
- MIT
- Primary implementation language
- JavaScript for the Gemini MCP bridge; orchestration and specialist definitions are Markdown
- Install surface
- Claude Code marketplace plugin with setup and uninstall slash commands
- Interface
- MCP over stdio, with Codex native MCP and a Gemini JSON-RPC bridge
- Provider requirement
- Requires at least one authenticated Codex CLI or Gemini CLI installation
- Runtime dependencies
- Gemini bridge uses built-in Node.js modules and declares itself zero-dependency
- Operating modes
- Read-only advisory mode and workspace-write implementation mode
- Specialist structure
- Five roles: architect, plan reviewer, scope analyst, code reviewer, and security analyst
Read from README.md, docs/plans/2026-02-13-multi-turn-and-config-defaults.md, docs/plans/2026-02-13-multi-turn-and-config-defaults-design.md, server/gemini/index.js, LICENSE, CLAUDE.md, CONTRIBUTING.md, commands/setup.md, rules/triggers.md, prompts/architect.md, commands/uninstall.md, config/providers.json, rules/orchestration.md, prompts/code-reviewer.md, prompts/plan-reviewer.md.
What it can do
Delegate code generation tasks to Codex
Natural language description of coding task → Generated code from Codex
Delegate tasks to Gemini AI
Task description or prompt → Response from Gemini AI
Switch between AI models during conversation
User selection of target AI model → Seamless handoff to selected model
Execute multi-model workflows
Complex task requiring different AI capabilities → Combined results from multiple AI models
Maintain session context across model switches
Ongoing conversation with model transitions → Continuous dialogue without context loss
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.