Agentic Coding

Using GLM 4.7 with Claude Code - Easy Setup with MoAI-ADK

01/05/2026

9.625min

Claude CodeGLM 4.7MoAI-ADKAgentic AIAI AgentsContext Engineering

A practical guide on connecting the GLM 4.7 model with Claude Code and automating configuration with MoAI-ADK's --glm-on command.

Using GLM 4.7 with Claude Code - Easy Setup with MoAI-ADK

Claude Code operates based on Anthropic's Claude models, but there are situations where leveraging the strengths of other models becomes necessary. This is particularly evident in large-scale codebase analysis or complex frontend tasks.

Z.AI's GLM 4.7 offers a 200K token context window and enhanced programming capabilities. This document covers how to integrate GLM 4.7 with Claude Code and explains the convenient setup method through MoAI-ADK's automation features.

1. Key Features of GLM 4.7

1.1 Core Technical Specifications

The most notable feature of GLM 4.7 is its 200K token context window. This resolves the issue of context loss in large-scale projects. It enables analyzing entire codebases in a single session.

The key strengths of GLM 4.7 are as follows:

Enhanced Programming Capabilities

Benchmark tests show high performance in complex refactoring tasks. It demonstrates excellent context maintenance during multi-stage requirement processing, and maintains consistency in code generation.

GLM-4.7 LLM Performance Evaluation - Benchmark Comparison

Large Code Processing

With 200K token context, most files in typical projects can be processed in a single session. Support for 128K token output ensures completion without interruption even during large code generation.

Transparent Reasoning Process

The interleaved reasoning feature exposes the model's thought process. This is useful for debugging tasks and enables verification of the model's conclusion derivation process.

Frontend Task Optimization

Shows improved performance in React component and CSS tasks. High quality in UI layout suggestions, and excellent ability to maintain style consistency.

1.2 Benefits of Claude Code Integration

The combination of Claude Code's agent-based workflow and GLM 4.7 offers the following advantages:

Single Processing of Complex Tasks

Large refactoring tasks can be handled in a single request without division. Efficiency improves for simultaneous frontend and backend modification tasks.

Enhanced Debugging Efficiency

Accuracy of error cause analysis increases through full context understanding. Verification of model judgments becomes easier due to exposed reasoning process.

Improved Development Speed

Time for repetitive tasks is reduced. Automation of test code writing and documentation tasks is effectively performed.

The chart below shows performance comparison in actual development scenarios published by Z.AI.

GLM-4.7 vs GLM-4.6 Real-world Development Scenarios - Performance Comparison in Real Development Environment

2. Manual Setup Method

2.1 Step-by-Step Setup Procedure

The manual method for connecting GLM 4.7 is as follows:

Step 1: Obtain Z.AI API Key

Create an account on the Z.AI website and obtain an API key.

Step 2: Modify settings.json File

The Claude Code configuration file is located at:

Bash

# Configuration file location
~/.claude/settings.json

Modify the configuration file as follows:

JSON
{
  "env": {
    "ANTHROPIC_AUTH_TOKEN": "your_zai_api_key",
    "ANTHROPIC_BASE_URL": "https://api.z.ai/api/anthropic",
    "API_TIMEOUT_MS": "3000000"
  }
}

Step 3: Restart Terminal

After saving the configuration, restart the terminal to apply changes.

2.2 Limitations of Manual Setup

The manual setup method has the following issues:

Path Navigation Complexity: The ~/.claude/ folder is set as a hidden directory, making access inconvenient
JSON Syntax Error Risk: Missing punctuation causes Claude Code startup to fail
Security Configuration Requirements: Separate permission settings for API key files are needed
Git Commit Risk: API keys may be included in version control if .gitignore is not set

Automation tools are needed to resolve these issues. MoAI-ADK, introduced in the next section, provides the solution.

3. Introduction to MoAI-ADK

Before addressing the complexity of manual setup, let me briefly introduce MoAI-ADK, the tool used in this document.

3.1 Project Overview

MoAI-ADK (Agentic Development Kit) is an open-source framework that integrates SPEC-First development methodology, Test-Driven Development (TDD), and AI agents. "MoAI" means "AI for Everyone," with the goal of realizing a transparent development lifecycle.

3.2 Core Features

MoAI-ADK provides 5 core features:

SPEC-First Methodology: Specification-based development using EARS (Easy Approach to Requirements Syntax) format is supported. Explicit requirement documentation reduces rework by approximately 90%.

TDD Automation: Automated Red-Green-Refactor cycles are applied. Minimum 85% test coverage is required, and bug occurrence rate is reduced by approximately 70%.

AI Agent Orchestration: Mr. Alfred coordinates over 24 specialized agents across 5 tiers. Development time is reduced by 60-70% for complex tasks.

Automatic Documentation: Changes are automatically synchronized to project documentation. 100% documentation currency rate is maintained.

TRUST 5 Quality Framework: Enterprise-grade quality is assured across 5 areas: test coverage, code readability, standard consistency, security checks, and change tracking.

3.3 Development Workflow

MoAI-ADK operates with a 3-stage cyclic workflow: Plan, Run, Sync.

Loading diagram...

MoAI-ADK Development Workflow - Plan, Run, Sync Cycle Structure

Plan: The /moai:1-plan "feature description" command generates specifications including success criteria and test scenarios.
Run: The /moai:2-run SPEC-001 command executes TDD implementation with automatic quality verification.
Sync: The /moai:3-sync SPEC-001 command automates documentation updates and prepares for deployment.

3.4 Installation Method

Installing MoAI-ADK is simple.

System Requirements:

Python 3.11-3.14 uv package manager

Installation Procedure:

Bash
# Install uv
curl -LsSf https://astral.sh/uv/install.sh | sh

# Install MoAI-ADK
uv tool install moai-adk

# Initialize project
moai-adk init my-project
cd my-project && claude

# Configure project
> /moai:0-project

MoAI-ADK is now ready. The next section covers the --glm-on feature that automates GLM 4.7 connection.

4. --glm-on Automation Feature

4.1 Feature Overview

The --glm-on feature of MoAI-ADK resolves the issues of manual setup discussed earlier. This feature handles the following tasks with a single command:

Automatic file path navigation
JSON syntax error prevention
Security configuration automation
Automatic .gitignore updates

4.2 Usage Method

Configuration is completed with a single command:

Bash

/moai:0-project --glm-on <your_api_token>

The tasks performed by this command are:

Automatic File Generation: The .env.glm file is automatically created and the API token is stored. .claude/settings.local.json configuration is also handled automatically. No need to manually write JSON structure.

Automated Security Configuration: .env.glm file permissions are set to 0o600, allowing only the owner to read/write. No need for separate chmod command execution.

Prevent Git Mistakes: .env.glm is automatically added to .gitignore. Accidental commits of API keys are prevented.

Configuration Verification: After command execution, verification of generated files is performed and results are displayed. Errors can be confirmed immediately if they occur.

Loading diagram...

GLM 4.7 Setup Sequence - Execution Flow of /moai:0-project --glm-on Command

4.3 Automatic Token Resolution Feature

Token input is only required once initially. Afterwards, it is automatically searched and used.

Loading diagram...

Token Resolution Flow - API Token Automatic Search Priority

Token Search Order:

Command Parameter: Token provided in --glm-on <token> format is used.
Stored File: Token is loaded from existing .env.glm file.
Environment Variable: System ANTHROPIC_AUTH_TOKEN environment variable is used.
User Input Request: If token cannot be found by above methods, input is requested.

After initial setup, token management is handled automatically.

4.4 Execution Examples

Initial Setup:

Execute the command with your API token:

Bash

/moai:0-project --glm-on sk-xxx-your-api-token-here

Execution result is displayed as follows:

Plain Text
✓ .env.glm file created (permissions: 0o600)
✓ .claude/settings.local.json configured
✓ .gitignore updated
✓ GLM 4.7 setup completed successfully!
> Restart Claude Code to apply new settings.

Subsequent Setup:

Token can be omitted:

Bash

/moai:0-project --glm-on

Generated .env.glm File Structure:

Bash
# GLM 4.7 API Configuration
# Generated by MoAI-ADK /moai:0-project --glm-on
# DO NOT commit this file to version control

ANTHROPIC_AUTH_TOKEN=sk-xxx-your-api-token-here
ANTHROPIC_BASE_URL=https://api.z.ai/api/anthropic
API_TIMEOUT_MS=3000000

Comments in the file clearly indicate the purpose of the configuration file.

4.5 Cost Efficiency

GLM integration through MoAI-ADK provides a cost-effective alternative. While Anthropic Claude Max costs $200 per month, Z.AI's GLM 4.7 offers equivalent coding performance at a much lower price.

Z.AI Monthly Subscription Plans:

Plan	Monthly Price	First Month Discount	Key Features
Lite	$6/month	$3 (50% off)	3x Claude Pro usage, for light tasks
Pro	$30/month	$15 (50% off)	5x Lite usage, 40-60% faster, Vision/Web Search support
Max	$60/month	$30 (50% off)	4x Pro usage, peak hour performance guarantee, early access to new features

Additional Discount for Annual Subscription (60% off):

Plan	Annual Price	First Year Discount	Monthly Equivalent
Lite	$72/year	$28.8 (60% off)	$2.4/month
Pro	$360/year	$144 (60% off)	$12/month
Max	$720/year	$288 (60% off)	$24/month

Cost Comparison with Claude Code Max:

Z.AI Max annual plan offers very high cost efficiency.

Service	Monthly Cost	Annual Cost	Notes
Claude Max	$200/month	$2,400/year	Anthropic official
Z.AI Max (annual)	$24/month equivalent	$288/year (first year)	88% savings
Z.AI Max (annual, regular)	$60/month equivalent	$720/year	70% savings

Based on the first year, Z.AI Max annual plan ($288) is cheaper than Claude Max monthly fee ($200). You can use the same 200K token context window and advanced coding features while saving $2,112 annually.

Seasonal Discount Benefits:

Z.AI offers additional discounts during Christmas season and Black Friday periods. Quarterly subscriptions get -55% discount, annual subscriptions get -60% discount, and during seasonal promotion periods, an additional 10-20% discount is applied.

Z.AI Yearly Pricing - Annual Subscription Season Discount Pricing

Referral Discount Benefits:

Signing up through the link below applies an additional 10% discount.

Referral credits generated from sign-ups through the above link are fully used for open-source project and R&D sponsorship. It operates in a way that contributes to the development of the open-source ecosystem, including MoAI-ADK.

Conclusion

This document covered the integration method of GLM 4.7 with Claude Code. The key points are summarized as follows:

Value of GLM 4.7: With a 200K token context window, single-session analysis of large-scale projects is possible. It provides enhanced quality for frontend tasks, and the interleaved reasoning feature enables verification of the model's thought process.

Limitations of Manual Setup: Risks exist such as JSON syntax errors, missing security configurations, and Git commit mistakes. This is a cumbersome task even for experienced developers.

MoAI-ADK's Solution: The --glm-on feature automates all configuration with a single command. Security settings and Git management are handled automatically, and the automatic token resolution feature eliminates repetitive input.

Recommended Next Steps:

Get API key at Z.AI subscription page (referral link for 10% discount)
Install MoAI-ADK: uv tool install moai-adk
Connect GLM 4.7: /moai:0-project --glm-on <your_api_token>
Restart Claude Code and start development

MoAI-ADK is published under the MIT license, and feedback and improvement suggestions are accepted through GitHub issues.

References: