Codex에서 GlobalGPT CLI를 설치하고 사용하는 방법

What GlobalGPT Adds to Codex

GlobalGPT CLI, the bundled Skills, and the MCP server are three parts of the same integration. The CLI supplies the commands, the Skills teach Codex when and how to use them, and MCP exposes GlobalGPT capabilities as callable tools.

You do not need to choose only one of the three. Install the CLI first, then add the Skills and MCP connection so Codex can use the same GlobalGPT account in different ways.

전제 조건

Before starting, make sure you have:

• Codex installed and working.
• Node.js and npm available on your computer.
• A GlobalGPT account with enough credits for the models you plan to use.
• A browser for the one-time authorization step.

The npm installation command is the same on macOS, Windows, and Linux. You can type it in your normal terminal or give the command to Codex.

Install GlobalGPT CLI

Choose either method below. Both install the same official GlobalGPT CLI package and lead to the same setup steps afterward.

Method 1: Install GlobalGPT CLI in Terminal

Open Terminal, PowerShell, or your preferred command-line application and run the official installation command:

npm i -g @glbgpt/cli

Wait for npm to finish. A successful installation ends without an npm error and reports the packages that were added.

Verify the installed CLI:

glbgpt --version

Method 2: Ask Codex to Install GlobalGPT CLI

If you prefer to stay inside Codex, send this instruction:

Install the official GlobalGPT CLI with this command:

npm i -g @glbgpt/cli

After installation, verify the installed package version and confirm
that both glbgpt and globalgpt command entries are available.

Codex can run the npm command, inspect the installed package, and verify the command entries for you.

Log In to GlobalGPT

Start the browser authorization flow:

glbgpt login

You can also ask Codex:

Start the GlobalGPT CLI login process.

The CLI opens an authorization page in your browser. Confirm that you initiated the request on your own device, then select Authorize. Do not send your password, verification code, or authorization URL to an AI assistant.

Check the active account privately with:

glbgpt whoami

Install the GlobalGPT Skills in Codex

The official CLI bundles four Skills for agent applications:

• globalgpt for account access and general GlobalGPT tasks.
• globalgpt-coding for explicitly delegated coding work.
• globalgpt-image for image generation and editing.
• globalgpt-video for video and image-to-video generation.

Install all four with:

glbgpt skill install

Once loaded, you can make a natural-language request such as:

Use GlobalGPT to research three recent productivity app launches
from official company blogs. Include the title, date, URL, and one
launch-message pattern from each source.

Connect GlobalGPT MCP to Codex

The MCP connection exposes GlobalGPT chat, model discovery, image generation, video generation, and task-status functions directly to Codex.

만약 glbgpt is available in your normal command path, register the stdio server with:

codex mcp add globalgpt -- glbgpt mcp --max-credits 4000

그리고 --max-credits value is a session budget for media generation. Registering the MCP server does not itself generate media or spend credits.

You can also ask Codex to configure it:

Register GlobalGPT as a Codex MCP server using:

glbgpt mcp --max-credits 4000

Then verify that the server is enabled. Do not generate any media yet.

Verify the saved configuration:

codex mcp get globalgpt

codex mcp list

• Chat with a GlobalGPT model.
• List chat, image, and video models.
• Generate or edit an image.
• Generate a video from text or reference frames.
• Check an image or video generation task.

Build a Complete AI Workflow in Codex

Installation only proves that the commands exist. The more useful question is what the integration adds to real work. To test that, we used GlobalGPT inside Codex to prepare a launch package for a fictional productivity app called FlowNote.

The workflow connected four different capabilities:

1. Use a web-enabled LLM to collect current official examples.
2. Turn the research into a launch brief.
3. Generate a landing-page hero image.
4. Animate the visual into an eight-second promotional video.

Step 1: Research Current Information with a Web-Enabled LLM

Tool-capable GlobalGPT chat models can search the web and fetch pages when a question requires current information. The test prompt asked for three recent productivity-product announcements from official company blogs:

Use web search to find three recent productivity app launches or
major product updates announced on official product blogs.

For each example, provide:
- Product name
- Exact announcement title
- Publication date
- Official source URL
- One landing-page or launch-message pattern worth learning from

Use official product or company sources, not news summaries.
Clearly separate verified facts from your recommendations.

The first valid response found Asana, Linear, and Notion announcements. We then checked each URL independently. All three returned HTTP 200, and the page titles and publication dates matched the answer.

Step 2: Turn the Research into a Launch Brief

The next GlobalGPT MCP call transformed those verified patterns into a practical FlowNote brief. The model defined the audience, core message, three benefits, landing-page structure, visual direction, video concept, and CTA.

Step 3: Generate a Landing-Page Hero Image

The installed globalgpt-image Skill first estimated the job, then generated one 16:9 image with the account’s default image model. The CLI selected Nano Banana 2 and estimated approximately 400 credits.

Create a clean 16:9 hero image for FlowNote, a fictional productivity
app that turns scattered notes into organized daily plans. Show a
modern desktop workspace with disconnected note cards smoothly
converging into one focused daily plan. Use a refined blue-and-violet
palette, soft depth, calm professional lighting, and generous negative
space. No readable text, logos, brand marks, watermarks, people, or
device mockup frames.

The image succeeded at the main concept: scattered notes visibly converge toward an organized planner in a blue-and-violet workspace. It did not follow every detail. The interface contains readable words such as “TODAY” and “Tasks,” despite the instruction to avoid readable text. We kept the first valid result instead of rerolling it.

If you want to refine your own prompt, see the 나노 바나나 2 프롬프트 가이드. You can also access Nano Banana 2 in GlobalGPT outside the CLI workflow.

Step 4: Turn the Hero Image into a Promo Video

그리고 globalgpt-video Skill used the hero image as the starting frame. The default video route selected Grok Imagine, estimated approximately 800 credits, and returned one eight-second video.

Animate the FlowNote hero image into an 8-second product-launch visual.
Disconnected paper notes and digital task cards should glide smoothly
from left to right and organize themselves into the focused daily
planner on the monitor. Use calm professional motion, subtle blue-and-
violet light trails, a locked camera, and a polished landing-page style.
Preserve the workspace composition. Do not introduce people, logos,
brand marks, watermarks, new readable text, camera shake, or abrupt
transitions.

The returned file was 1280×720, 8.042 seconds long, and encoded with H.264. A representative frame kept the desk, monitor, and note-card concept, but the composition changed substantially and the monitor interface became nearly blank. This is a useful reminder that image-to-video generation may reinterpret a reference frame rather than preserve it exactly.

You can explore the same model through Grok Imagine in GlobalGPT.

What This Workflow Demonstrates

The value is not any single answer or media file. Codex can coordinate the sequence: current web research becomes a brief, the brief becomes a visual direction, and the visual becomes a video. GlobalGPT supplies additional models and media tools while Codex keeps the project context and organizes the work.

Useful GlobalGPT CLI Commands

Use this compact reference when you need to check an account, discover a model, or start a generation task.

문제 해결

glbgpt: command not found

Close and reopen your terminal first. If the error remains, check whether npm’s global binary directory is included in your shell PATH. Installing the npm package and making the command discoverable are separate steps. You can also ask Codex to locate the installed package and explain which PATH entry is missing.

Codex does not see the newly installed Skills

Run the Skill installer, then restart Codex or open a new task. Skills are loaded when the agent starts; an already-open task may not refresh its Skill inventory automatically.

glbgpt skill install

The MCP server is enabled but no tools appear

Confirm the saved MCP configuration, then open a new Codex task:

codex mcp get globalgpt

If Codex starts outside your terminal environment, locate the full GlobalGPT executable path:

command -v glbgpt

A model returns no output

Treat authentication, transport, provider capacity, and empty responses as operational errors, not proof that the model is weak. Do not keep retrying media generation blindly because an interrupted server-side task may still consume credits. Check the error and task status first.

자주 묻는 질문

Are GlobalGPT CLI, Skill, and MCP the same thing?

They belong to the same integration, but they have different roles. The CLI supplies commands, the Skills guide Codex’s behavior, and MCP exposes functions as tools.

Can Codex install GlobalGPT CLI for me?

Yes. Give Codex the official command below and ask it to verify the installed version and command entries.

npm i -g @glbgpt/cli

Do I need to keep Terminal open?

No. After installation and configuration, Codex can launch the CLI or MCP server when needed. Terminal is useful for setup, verification, and troubleshooting.

Can GlobalGPT models search the web from Codex?

Tool-capable chat models can search the web and fetch pages. Web search may consume additional credits. Always verify important URLs, dates, prices, and product claims against the cited source.

Can I generate images and videos inside Codex?

Yes. The bundled globalgpt-image 그리고 globalgpt-video Skills use your GlobalGPT account, and the MCP server also exposes image and video tools.

Does MCP replace the CLI?

No. The MCP server is launched by the CLI. MCP makes the capabilities easier for Codex to call as tools, while the CLI remains the underlying local interface.

How do I switch GlobalGPT accounts?

How do I update GlobalGPT CLI?

Install the current package again:

npm i -g @glbgpt/cli

Refresh the bundled Skills, then restart Codex:

glbgpt skill install

최종 결론

GlobalGPT CLI turns Codex into a broader multi-model workspace. The CLI handles the local connection, Skills make natural-language delegation predictable, and MCP lets Codex call chat and media tools directly. Once all three are connected, a task can move from current research to structured planning, image generation, and video generation without breaking the workflow into unrelated apps.

If you want to use the same multi-model account outside the command line, you can also open GlobalGPT in your browser.