CCGLM

The CCGLM MCP Server lets Claude Code invoke another Claude Code instance that uses Z.AI’s GLM backend, all from your main Claude Code environment. It routes prompts through a dedicated MCP tool to the GLM-backed model, enabling code generation, analysis, and general queries without altering your primary settings.

How to use

You interact with the server through the MCP tool named glm_route. Use this tool to send your prompt and receive responses from the GLM-4.6 backend via Claude CLI. You can perform code generation, deep analysis, and general inquiries. When you want to run it from Claude Code, call the MCP tool with your prompt. You can also use the hashtag style commands to trigger specific tasks.

How to install

Prerequisites: Python and pip must be available on your system.

1. Install dependencies.

cd ~/IA/ccglm-mcp
pip install -r requirements.txt

Additional configuration and usage details

Register the MCP server by adding the following to your Claude settings so the MCP framework can start the local server.

{
  "mcpServers": {
    "ccglm_mcp": {
      "command": "python3",
      "args": ["/home/manu/IA/ccglm-mcp/ccglm_mcp_server.py"],
      "timeout": 300000
    }
  }
}

Optional alias and environment setup

Optionally, configure an alias to simplify usage and pass credentials automatically.

# Z.AI GLM Integration
export GLM_API_TOKEN="5951e8ffb37b4f038ba9a47f49e86ff4.ZYOeeSK1MWARJ8YB"
alias ccglm='ANTHROPIC_BASE_URL=https://api.z.ai/api/anthropic ANTHROPIC_AUTH_TOKEN=$GLM_API_TOKEN claude --dangerously-skip-permissions -c'

Recharge your shell with:

source ~/.zshrc

Testing the server

Run the test suite to verify file tracking, log sanitization, and basic prompt handling.

cd ~/IA/ccglm-mcp
python3 test_server.py

Security and logging

Keep the GLM credentials in a protected environment file and ensure logs sanitize tokens. Logs are sent to stderr to avoid interfering with the MCP protocol.

Configuration notes

Timeouts are defined in the server code with a default of 600 seconds and a maximum of 2400 seconds. Claude CLI runs with specific flags to streamline non-interactive usage.

Troubleshooting

If the Claude CLI is not found, ensure the Claude CLI is installed and accessible in your PATH.

If GLM authentication fails, verify the token in your environment and its active status.

For timeouts, adjust the server timeouts in the configuration file and restart Claude Code.

Usage from the terminal

Beyond the MCP, you can invoke GLM directly from the terminal using the configured alias.

# Interactive use
ccglm

# Prompt-based use
echo "Explain recursion" | ccglm

Updating

To update the server, pull the latest changes and reinstall dependencies, then restart Claude Code to apply the updates.

cd ~/IA/ccglm-mcp
git pull  # if this directory is a git repo
pip install -r requirements.txt --upgrade

Restart Claude Code to reload the MCP server."

## Notes

This server is designed to route prompts to the GLM-4.6 backend via Claude CLI and the Z.AI GLM API, using your main Claude Code environment without changing its core settings.

## Available tools

### glm\_route

Enables routing of prompts to GLM-4.6 via Claude CLI, supporting code generation, deep analysis, and general queries while excluding internal project files from tracking.