A bridge for delegating non-trivial software engineering tasks to a local Codex installation through the Model Context Protocol.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install codex-mcp-dev
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install codex-mcp-dev using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Codex MCP Dev?
Codex MCP Dev is a high-performance developer skill designed to integrate local Codex CLI capabilities directly into your AI agent environment. By leveraging the mcporter and codex-mcp-server stack, it provides a stable and reliable path for executing deep coding tasks. This integration is part of the broader ecosystem of Openclaw Skills, enabling agents to operate with project-wide context to build features, fix bugs, and refactor code.
Unlike simple text generation, this skill allows for machine-local execution, meaning the AI can interact with your file system and local development tools with high precision. It is specifically built for developers who need to bridge the gap between AI reasoning and local repository management, ensuring that complex development work is handled through a structured, protocol-compliant pipeline.
Codex MCP Dev Use Cases
- Implementing complex features across multiple files within a local repository.
- Automating the refactoring of legacy code while adhering to strict architectural constraints.
- Debugging failing test suites by providing the AI with real-time stack traces and repo context.
- Writing and updating unit or integration tests to improve codebase coverage.
- Performing read-only architectural analysis and root cause identification for reported bugs.
How Codex MCP Dev Works
- The user initiates a request for a non-trivial coding task through an agent using Openclaw Skills.
- The skill invokes the local Codex installation via a specialized Python wrapper script located in the project scripts directory.
- A health check or 'doctor' command is optionally run to ensure the MCP server and CLI dependencies are properly synced.
- The user selects a mode—either 'Analysis' for planning or '--full-auto' for direct implementation and file modification.
- Codex processes the prompt using the specified working directory, applying changes or generating insights based on project files.
- The developer inspects the output, validates changes with local tests, and finalizes the implementation session.
Codex MCP Dev Setup
To get started with this entry in the Openclaw Skills library, ensure you have mcporter and codex-mcp installed locally. Then, configure your project scope:
# Add the codex-cli MCP server to your project configuration
mcporter config add codex-cli --scope project --command codex-mcp
# Verify the installation and connection health
python3 scripts/codex_mcp.py doctor
Codex MCP Dev Data Schema & Taxonomy
The skill manages data primarily through filesystem interactions and command-line arguments to maintain project integrity:
| Data Type | Description |
|---|---|
| --cwd | Absolute path to the target repository or project directory |
| --prompt | String containing specific instructions, constraints, and acceptance criteria |
| --prompt-file | Path to a local file for handling large-scale context or complex instructions |
| --sandbox-mode | Security setting (e.g., read-only) to restrict file system modifications |
| --approval-policy | Defines when the agent requires manual confirmation before proceeding |
Codex MCP Dev Advanced Features
- Support for --full-auto mode for autonomous end-to-end implementation and test generation.
- Sandbox-mode configurations to allow for safe, read-only repository analysis without accidental edits.
- Integration with project-level mcporter.json for scoped, repo-specific AI agent behavior.
- Capability to handle long-form prompts via external file references for high-complexity engineering tasks.
- Specialized doctor commands for diagnostic troubleshooting of the MCP communication bridge.
SKILL.md
Loading
Related Openclaw Skills
Family Intent Recognition
An AI-powered classifier that transforms informal family chat logs into structured JSON data highlighting consumer needs and purchasing intent.
Nano Banana 2 Image Generation
A high-performance image generation and editing skill powered by Google's Nano Banana 2 model for creating professional visual assets.
Nano Banana Pro Image Generation & Editing
A premium AI skill for generating and editing high-fidelity professional images with advanced reasoning and precise text rendering.
Credit
A comprehensive AI agent skill for decoding credit score mechanics, building credit history, and disputing report errors.
Chile
A high-precision travel planning assistant for navigating Chile's complex geography, entry regulations, and macro-region logistics.
bibi
A command-line interface for BibiGPT that generates AI summaries of videos, audio, and podcasts directly in your terminal.
Featured*
