Claude Bridge Server

Table of Contents

• Why This Exists
• Quick Example
• Features
• When NOT to Use This
• Prerequisites
• Quick Start
• Documentation
• Works with Other MCP Clients Too
• Security
• License

---

Why This Exists

You have a Claude subscription but you're hitting limits. Getting a second Claude subscription is pricey (or upgrading might be too expensive) — but a Z.ai subscription costs less and GLM codes just as well as Sonnet.

The question: How do you make Claude and GLM work together?

The answer: Claude Bridge. An MCP server that coordinates both through a shared task queue.

┌─────────────────────────┐                    ┌─────────────────────────┐
│   Terminal 1            │                    │   Terminal 2            │
│   Claude Opus           │                    │   Z.ai GLM              │
│   (Architect)           │                    │   (Executor)            │
│                         │    Claude Bridge   │                         │
│   • Plans features      │◄──────────────────►│   • Pulls tasks         │
│   • Designs architecture│    (Shared Queue)  │   • Writes code         │
│   • Makes decisions     │                    │   • Reports back        │
└─────────────────────────┘                    └─────────────────────────┘

Quick Example

Terminal 1 (Opus): "Push a task to implement JWT authentication"

Terminal 2 (GLM): "Pull the next task" → implements → "Complete the task"

Terminal 1: "What did the executor complete?" → reviews results

Features

• Task Queue — Priority-based with dependencies and categories
• Shared State — Project focus and decisions visible to both
• Session Context — Resume where you left off
• Clarifications — Executor asks questions, Architect responds
• Token-Conscious — Designed to minimize Opus usage

When NOT to Use This

If you're using a planning framework like Get-Shit-Done (GSD), you don't need this bridge. Frameworks like GSD store context in project files (.planning/) that both terminals can read directly. Slash commands like /gsd:plan-phase and /gsd:execute-phase work in any terminal.

Use Claude Bridge for:

• Ad-hoc tasks, quick fixes, one-off research
• Projects without a planning framework
• Session continuity across restarts

Skip the bridge if:

• Using GSD or similar file-based planning frameworks
• Both terminals share the project filesystem

Prerequisites

• Node.js 18+ — Download
• Claude Code CLI — Installation
• Z.ai account — For GLM access (z.ai)
• Two terminal windows

Terminal Profile Setup

Same subscription in both terminals? Just run claude in each — no profile setup needed.

Claude + Z.ai GLM? Add this function to your shell profile:

Windows PowerShell (notepad $PROFILE):

function use-glm {
    $env:ANTHROPIC_BASE_URL = "https://api.z.ai/api/anthropic"
    $env:ANTHROPIC_AUTH_TOKEN = "your_zai_api_key_here"
    Write-Host ">>> GLM ACTIVE (Executor mode)" -ForegroundColor Cyan
}

macOS/Linux (~/.bashrc or ~/.zshrc):

use-glm() {
    export ANTHROPIC_BASE_URL="https://api.z.ai/api/anthropic"
    export ANTHROPIC_AUTH_TOKEN="your_zai_api_key_here"
    echo ">>> GLM ACTIVE (Executor mode)"
}

Usage:

• Terminal 1: Just run claude (uses your Claude subscription)
• Terminal 2: Run use-glm, then claude (uses Z.ai API)

See GUIDE.md for complete setup details.

Quick Start

# 1. Clone and build
git clone https://github.com/Joncik91/claude-bridge-server.git
cd claude-bridge-server
./setup.sh          # or setup.bat on Windows

# 2. Configure MCP (see GUIDE.md for details)
claude mcp add --scope user --transport stdio claude-bridge -- node /path/to/server/dist/index.js

Documentation

GUIDE.md — Complete setup, terminal configuration, usage, and troubleshooting

Works with Other MCP Clients Too

While designed for Claude + GLM, the bridge is a standard MCP server — it works with any MCP-compatible CLI:

• Two Claude Code instances (same or different subscriptions)
• Gemini CLI
• OpenCode
• Qwen Code
• Any other MCP client

Security

Claude Bridge sits between two CLIs that hold real API tokens. Three implications worth being explicit about.

• API tokens never pass through the bridge. The server's protocol carries tasks, statuses, and results — not credentials. Each terminal authenticates against its own provider directly (your Claude subscription on one side, your Z.ai key on the other). The shell-profile pattern in Prerequisites is there so the terminal session sets ANTHROPIC_AUTH_TOKEN — the bridge never reads it.
• Don't commit use-glm with a literal key. The shell-profile snippets in this README and GUIDE.md use a placeholder — replace it with ${ZAI_API_KEY} or similar and load the secret from a password manager / system keychain. A literal key in ~/.bashrc is a literal key on every backup of your home directory.
• The shared task queue is local to the bridge process. Tasks, clarifications, and reviewer notes live in the bridge's in-memory state. Don't aim two unrelated projects at the same bridge instance — the queue is single-tenant by design.

For team use: each pair of architect+executor terminals should run its own bridge process, not share one. The MCP transport is stdio on the local machine; nothing listens on a network port.

License

MIT — see LICENSE.

Acknowledgments

• Built for Claude Code by Anthropic
• Designed for Z.ai GLM integration