Discover MCPs & agents
Loading MCPs and agents…
Loading MCPs and agents…
Framework for running Claude Code + Gemini CLI as parallel AI coding assistants
From the repo.
⚠️ No longer maintained. This project has been superseded by a multi-model council workflow (multiple AI CLIs convened as reviewers from a single session), which replaced the split-terminal approach. The repo is archived and kept for reference.
Run two AI coding assistants in parallel — one builds, one thinks.
A framework for running Claude Code and Gemini CLI as parallel AI coding assistants on the same project. Each AI gets its own context file with a defined role, shared conventions, and instructions to sync with the other — so they stay aligned without stepping on each other.
Most AI coding workflows use one model. This uses two in split terminals:
The key is that both AIs read from shared context files that keep them in sync. When Gemini makes an architecture decision, it gets written down. When Claude ships a feature, it gets written down. Neither AI works in a vacuum.
Context files follow three rules: short, opinionated, operational.
Run new-project.sh and it creates three files in your project:
./new-project.sh my-app "Next.js, Supabase, Vercel"
CLAUDE.md — Claude Code's context fileLoaded automatically by Claude Code at session start. Contains:
GEMINI.md — Gemini CLI's context fileLoaded by Gemini CLI. Contains:
ROUTING-PROTOCOL.md — Decision frameworkA reference doc (for you, the developer) that defines when to use which AI:
git clone https://github.com/OSideMedia/dual-brain.git
cd dual-brain
chmod +x new-project.sh
./new-project.sh my-app "Next.js, Supabase, Vercel"
Then:
cd ~/Projects/my-app<!-- comments --> in CLAUDE.md and GEMINI.mdThe stack argument is optional — defaults to Next.js, Supabase, TypeScript, Tailwind CSS, Vercel. Pass whatever matches your project.
If you have existing projects using v1 templates, you can either re-scaffold or audit manually:
new-project.sh again, then copy your project-specific rules and decisions into the new structure./audit-claude-md slash command in Claude Code to automatically identify what stays and what moves to README/docs.npm install -g @anthropic-ai/claude-codenpm install -g @google/gemini-cliReplace {MCP_ENDPOINT_URL} with this MCP’s endpoint URL (from its repo or docs above). No API key — you connect directly.
Tool
OS
Config file: ~/.cursor/mcp.json
{
"mcpServers": {
"mcp-server": {
"url": "{MCP_ENDPOINT_URL}"
}
}
}Paste into mcpServers in the config file. Restart Cursor after saving.
If this MCP is also published on mcpchannel.ai, you can subscribe from Browse and use the gateway config there instead.