Frequently Asked Questions

General Questions

What is Codex MCP Tool?

Codex MCP Tool is a Model Context Protocol (MCP) server that bridges OpenAI's Codex CLI with MCP-compatible clients like Claude Desktop and Claude Code. It enables seamless AI-powered code analysis, generation, and brainstorming directly within your development environment.

Why use Codex MCP Tool instead of Codex CLI directly?

Benefits of using Codex MCP Tool:

• Non-interactive execution - No manual prompts or confirmations needed
• Integration - Works seamlessly with Claude Desktop/Code
• File references - Use @ syntax to include files easily
• Structured output - Get organized responses with changeMode
• Progress tracking - Real-time updates for long operations
• Multiple tools - Access specialized tools beyond basic prompts

Which AI models are supported?

By default the server omits model and Codex CLI uses the model from ~/.codex/config.toml. Known models (Codex CLI 0.159.2):

• gpt-6.1-sol - Latest workhorse for coding and everyday work (Codex CLI 0.159+)
• gpt-6-astra - Frontier intelligence for the most demanding work
• gpt-6-sol - Previous-generation workhorse
• gpt-6-luna - Fast and affordable; easier, repeatable tasks
• gpt-5.6-sol / gpt-5.6-terra / gpt-5.6-luna - Older generation, still available
• gpt-5.5 - Legacy; Codex steers callers to gpt-5.6-sol

Use a concrete slug: the bare aliases gpt-6 and gpt-5.6 are rejected with HTTP 400.

Is this an official OpenAI tool?

No, this is a community-developed integration tool. It uses the official Codex CLI but is not directly affiliated with or endorsed by OpenAI.

Installation & Setup

What are the prerequisites?

1. Node.js >= 18.0.0
2. Codex CLI installed and authenticated
3. MCP client (Claude Desktop or Claude Code)
4. OpenAI API access with appropriate model permissions

How do I install Codex MCP Tool?

For Claude Code (recommended):

claude mcp add codex-cli -- npx -y @trishchuk/codex-mcp-tool

For Claude Desktop:

npm install -g @trishchuk/codex-mcp-tool

Then add to your configuration file.

Where is the configuration file located?

Claude Desktop configuration locations:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json
• Linux: ~/.config/claude/claude_desktop_config.json

How do I verify the installation?

Test with these commands in your MCP client:

// Test connectivity
/codex-cli:ping "Hello"

// Check help
/codex-cli:help

// Simple task
/codex-cli:ask-codex "explain what Python decorators are"

Usage Questions

How do I reference files in my prompts?

Use the @ syntax:

// Single file
'explain @src/main.ts';

// Multiple files
'compare @src/old.ts @src/new.ts';

// Glob patterns
'review @src/*.ts';
'analyze @src/**/*.ts';

// With paths containing spaces (use quotes)
"@\"My Documents/project/file.ts\"";

What's the difference between sandbox modes?

Mode	Read	Write	Delete	Execute	Use Case
read-only	✅	❌	❌	❌	Analysis, reviews
workspace-write	✅	✅	⚠️	❌	Refactoring, generation
danger-full-access	✅	✅	✅	✅	Full automation

How do I use different models?

Specify the model in your request:

{
  "name": "ask-codex",
  "arguments": {
    "prompt": "your task",
    "model": "gpt-6-astra"  // or "gpt-6.1-sol", "gpt-6-luna"
  }
}

What is changeMode?

ChangeMode returns structured file edits instead of conversational responses:

{
  "prompt": "refactor this code",
  "changeMode": true
}
// Returns OLD/NEW edit blocks that can be directly applied

How do I handle large responses?

For large changeMode responses, use chunking:

// Initial request returns cacheKey
{ "prompt": "large refactor", "changeMode": true }

// Fetch subsequent chunks
{ "cacheKey": "abc123", "chunkIndex": 2 }

Tools & Features

What tools are available?

1. ask-codex - Execute Codex commands with file references
2. brainstorm - Generate ideas with structured methodologies
3. ping - Test connectivity
4. help - Show Codex CLI help
5. fetch-chunk - Retrieve cached response chunks
6. timeout-test - Test long-running operations

How does the brainstorm tool work?

The brainstorm tool offers multiple methodologies:

{
  "prompt": "ways to improve performance",
  "methodology": "scamper",  // or "divergent", "convergent", etc.
  "domain": "backend",
  "ideaCount": 10,
  "includeAnalysis": true
}

Can I create custom tools?

Yes! Create a new tool in src/tools/:

// src/tools/my-tool.tool.ts
export const myTool: UnifiedTool = {
  name: 'my-tool',
  description: 'My custom tool',
  schema: MyToolSchema,
  async execute(args, progress) {
    // Implementation
  },
};

How do progress notifications work?

Long-running operations send progress updates every 25 seconds:

// In tool implementation
progress?.('Processing file 1 of 10...');
progress?.('Analyzing dependencies...');
progress?.('Generating output...');

Security & Privacy

Is my code sent to OpenAI?

Yes, when you use Codex MCP Tool, your prompts and referenced files are sent to OpenAI's API for processing. Ensure you:

• Don't include sensitive data
• Review OpenAI's data usage policies
• Use appropriate sandbox modes

How do approval policies work?

Policy	Description	Use Case
never	No approvals needed	Trusted automation
on-request	Approve each action	Careful operation
on-failure	Approve on errors	Semi-automated
untrusted	Always require approval	Maximum safety

Can I use this in production?

Codex MCP Tool is designed for development environments. For production:

• Review security implications
• Implement proper access controls
• Monitor API usage and costs
• Consider rate limiting

Are API keys stored securely?

API keys should be:

• Set via environment variables (OPENAI_API_KEY)
• Never committed to version control
• Managed through Codex CLI authentication
• Rotated regularly

Troubleshooting

Why is the tool not responding?

Check:

1. Codex CLI is installed: codex --version
2. Authentication is valid: codex auth status
3. MCP server is running: restart your client
4. Configuration syntax is correct

How do I enable debug logging?

# Enable all debug output
DEBUG=* npx @trishchuk/codex-mcp-tool

# Enable specific modules
DEBUG=codex-mcp:* npx @trishchuk/codex-mcp-tool

What if I get "model not available"?

1. Check available models: codex models list
2. Verify API access permissions
3. Try a different model (e.g., gpt-6-luna)
4. Check OpenAI account status

How do I report bugs?

1. Check existing issues
2. Gather diagnostic information
3. Use the bug report template
4. Include reproducible steps

Advanced Topics

Can I use multiple MCP servers simultaneously?

Yes, configure multiple servers in your MCP client:

{
  "mcpServers": {
    "codex-cli": { ... },
    "another-server": { ... }
  }
}

How do I contribute to the project?

See our Contributing Guide:

1. Fork the repository
2. Create a feature branch
3. Make your changes
4. Submit a pull request

What's the difference between Codex models?

• Astra - the most capable tier, for hard or open-ended work
• Sol - the everyday workhorse; gpt-6.1-sol is the newest
• Luna - the fast, affordable tier for well-specified tasks; tops out at max reasoning
• Terra (GPT-5.6 only) - balanced capability and cost

Can I use this with other AI providers?

Currently, Codex MCP Tool is designed for OpenAI's models via Codex CLI. For other providers, consider:

• Forking and adapting the codebase
• Using different MCP servers
• Contributing multi-provider support

Performance & Optimization

How can I improve response times?

1. Use faster models: gpt-6-luna with reasoningEffort: "low"
2. Be specific with file references: Avoid broad globs
3. Enable caching: Reuse common analyses
4. Process in batches: Break large tasks

What are the context limits?

Model	Context window	Max context window
gpt-6.1-sol, gpt-6-astra/sol/luna	272K	872K
gpt-5.6-sol, gpt-5.6-terra, gpt-5.6-luna	272K	872K
gpt-5.5	272K	272K

Values come from Codex's models_cache.json (client 0.159.2).

How do I estimate costs?

// Rough cost calculation
tokens = prompt_length + file_content_length + response_length;
cost = (tokens / 1000) * model_price_per_1k;

// Model prices vary - check OpenAI pricing for current rates
// Astra: highest cost; Sol: everyday; Luna: most economical

Future & Roadmap

What features are planned?

• Streaming responses
• Local model support
• Enhanced caching
• Web UI interface
• Custom model configurations
• Batch processing
• Team collaboration features

How do I request features?

1. Check existing requests
2. Use the feature request template
3. Provide use cases and examples

Is there a roadmap?

Check our GitHub Projects for planned features and progress.

Related Resources

• Installation Guide
• Troubleshooting Guide
• API Reference
• Examples
• Contributing