Google Jules × Claude Code × Codex

Run dozens of Jules agents from one command.

jules-dispatch orchestrates Google Jules in parallel — declarative tasks, live monitoring, and an MCP server so Claude Code and Codex can drive it directly.

View on GitHub
✓ MIT licensed ✓ Node.js 20+ ✓ Production-ready
TS TypeScript · Node.js
MCP Model Context Protocol
Works with Claude Code · OpenAI Codex CLI
Any OpenAI-compatible LLM (OpenAI / Ollama / Groq / OpenRouter)
The problem

Jules is powerful. One task at a time is not.

Jules runs a single task per turn. When you're migrating a framework, refactoring a monolith, or adding types across 20 modules, serial execution becomes the bottleneck.

✕ Without jules-dispatch
  • ✕Write the same prompt by hand, over and over
  • ✕Wait for one task to finish before the next starts
  • ✕No way to scale a refactor across a codebase
  • ✕AI agents can't call Jules in parallel on their own
↗ With jules-dispatch
  • ✓Describe a goal in one line; an LLM splits it into parallel tasks
  • ✓Launch 10–50 Jules sessions from a single command
  • ✓Tail logs, approve plans, message, and cancel — all in real time
  • ✓Claude and Codex treat Jules as a first-class tool
What it does

Built for parallel, AI-assisted development.

Each capability targets a real bottleneck in large-scale AI coding.

01

Parallel scheduling

Launch dozens of sessions with --parallel 20. Smart queueing keeps you under rate limits.

Up to 50+ concurrent
02

Declarative YAML

Define tasks as code: title, repo, branch, prompt. Multi-document files, version control, team-ready.

03

Autonomous AI planning NEW

Describe an intent in one line; the built-in LLM planner breaks it into parallel tasks. Any OpenAI-compatible model.

04

Lifecycle control

Live tail log streams, plan approval, follow-up messages, cancel runaway sessions, smart retry with backoff. Human-in-the-loop or fully autonomous.

05

Native MCP server

Run jules-dispatch mcp to expose Jules as a tool for Claude Code and Codex. Your agent now has real parallel execution.

12 tools exposed
06

Built for agents and CI

--json structured output, audit logs, doctor diagnostics, per-call API keys, stdin support. Designed for scripts, CI, and agents.

How it works

From intent to pull request.

A few steps, fully observable.

01
Define
Describe the work in YAML. Batch multiple tasks per file.
02
Dispatch
jules-dispatch fans out, rate-limits, retries, and tracks state.
03
Collect
Detect PRs, plan approvals, and failures. Live tail and structured output.
New in v1.2

Stop writing tasks by hand. Let an LLM plan them.

Give jules-dispatch one sentence and it calls your configured LLM (OpenAI / Ollama / Groq / OpenRouter) to split it into parallel Jules tasks.

bash
jules-dispatch auto "Migrate every Express route to Fastify, and add request-validation tests for each."

It produces 6 independent tasks — auth routes, user routes, billing routes, middleware, the bootstrap file, and tests — then asks whether to dispatch them in parallel.

Supported LLM providers
OpenAI (default)
Ollama (local)
Groq
OpenRouter
Together / Fireworks / DeepInfra / Azure OpenAI / LiteLLM — any compatible endpoint
Configure via environment variables or flags. No code changes.
Get started in 4 minutes

Four steps to parallel.

01
Global install
Install once, use everywhere
bash
npm install -g @yuuqq/jules-dispatch
02
Initialize
Connect your Jules API key
bash
jules-dispatch init

Interactive wizard sets JULES_API_KEY. Pass --api-key for a single run.

03
Declare a task (YAML)
Version-controlled, shareable

Create a tasks/ directory and define tasks in simple YAML (multi-document --- supported):

yaml
# tasks/migrate-to-fastify.yaml
title: "Migrate auth routes to Fastify"
repo: "your-org/your-repo"
branch: "main"
prompt: |
  Migrate every Express route under /api/auth to Fastify.
  Use the Fastify plugin system; preserve existing middleware.
  Add schema validation where appropriate.
04
Dispatch in parallel
Start your swarm
Single · batch · AI-planned
bash
jules-dispatch dispatch tasks/migrate-to-fastify.yaml
bash
jules-dispatch batch tasks/ --parallel 12
bash
jules-dispatch auto "Add full TypeScript types and Vitest tests across the project"
Full command reference & advanced options ↗
For AI agents

Give your coding agent real parallel execution.

jules-dispatch isn't just a CLI — it's an MCP server.

  • ▸Claude Code can call jules_dispatch and friends directly
  • ▸Codex CLI integrates the same way
  • ▸12 tools exposed: dispatch, wait, approve, cancel, plan, and more
Configure in Claude Code (example)
json
{
  "mcpServers": {
    "jules": {
      "command": "jules-dispatch",
      "args": ["mcp"],
      "env": {
        "JULES_API_KEY": "your-jules-api-key-here",
        "LLM_API_KEY": "sk-..."   // optional, for AI planning
      }
    }
  }
}

Once configured, just say in Claude Code: Use Jules to refactor the user and order modules in parallel.

See MCP-INTEGRATION.md or CLAUDE.md for details.

Ready to put Jules to work in parallel?

jules-dispatch is the simplest way to orchestrate Google Jules at scale.

Star on GitHub
MIT License · Built by developers, for developers.