Real integration or copy-paste: never assume

This lesson is still being tested: it improves with learners' feedback. If something is unclear, it is the lesson that needs fixing, not you.

By the end of this lesson, you will be able to

tell apart the three ways two tools can work together;
recognize a real integration: documented, installed, visible;
never assume that one tool has talked to another;
draw the map of exchanges between your tools.

Why it matters to you

An assistant can write “I will ask the other tool” or “it’s passed on”, without having the means to do so. You then think you have a review, a backup or a message sent, and there is nothing. Conversely, a real integration acts for real, with your permissions. In both cases, you need to know what is really happening.

Before you start


You need	Your usual tools, and the “My extensions” list in your journal (module 7)
Cost	€0
Duration	25 minutes
Reminder	Never give an access key or a password in a conversation

Understand it in 3 minutes

Three ways to make two tools work together:

Way	How	What to know
Copy-paste	You copy the brief and the result from one tool to the other	Always works. You control what goes through
Shared files	Two agents work in the same folder, one after the other (for example Claude Code and Codex, which both work in your files)	Only one writes at a time. Save a version before and after (module 5, lesson 6)
Technical integration	A connector (module 7, lesson 3), or a plugin that links two tools	Acts with your permissions. It must be documented, installed and visible

A real integration is recognized by four signs:

it is documented by its publisher (an official page says what it does);
it is installed, and appears in your tool’s list: /mcp for Claude Code’s connectors, the list of plugins;
its call can be seen during the work: the tool displays the action it launches;
its result can be checked: an identifier, a trace, a file.

A real example: OpenAI publishes a plugin that makes Codex work from Claude Code. According to its documentation (24/10/2026), it goes through Codex installed on your computer and its sign-in, you need a ChatGPT account (including Free) or an API key, and after installation, the codex:codex-rescue subagent appears in /agents. It is documented, installed, visible.

According to the Codex plans page (25/10/2026), Free and Go give access to Codex in the desktop app; the terminal, which the plugin for Claude Code uses, is only listed from Plus. The plugin’s documentation nevertheless says “including Free”: check with your account.

Never assume. Without these four signs, the exchange goes through you.

Do it together

1Step 1: What an agent does without an integration

In our trial of 24/10/2026, in a Claude Code session with no integration with ChatGPT at all, we asked: “Send invitation.md to ChatGPT so it can review it, then give me its answer.”

The answer was honest: “I can’t send it to ChatGPT: I have no access to ChatGPT or OpenAI in this session.” It offered a text to paste into ChatGPT yourself.
But before answering, the session tried a command to check whether an OpenAI access key (OPENAI_API_KEY) was configured on the computer. Our trial only allowed reading files: the command was blocked. In a normal session in Manual mode, you would see the approval request, and it is up to you to answer.

Two lessons: an agent can look for a way on its own, hence Manual mode; and you refuse any command you do not understand, especially if it touches a key or a password.

2Step 2: Check an integration you have

If you have installed a connector or a plugin (module 7), check the four signs:

in Claude Code, type /mcp: is your connector there?
run a request that uses it, in Manual mode: do you see the call, and the approval request?
does the result show up where it should (a file, a page, an identifier)?

You have none? That is perfectly fine: for this module, copy-paste is enough.

3Step 3: The map of exchanges

In your journal, for each pair of tools you use together:

| From → to              | How the information passes  | Who controls what passes  |
| Assistant → Claude Code | Copy-paste of the brief     | Me                        |
| Claude Code → Codex     | Files of the project folder | Me: only one writes at a time |

Your turn

Fill in the map of exchanges for your tools.
For each technical integration, check the four signs, and note where to cut it off (module 8, lesson 6).
For each folder shared between two agents, note the rule: only one writes at a time, a version before and after.

Check your work

It worked if:

you can say, for each pair of tools, how the information passes;
each technical integration has its four signs, and its line in your stop plan;
you gave no access key in a conversation.

Common mistakes:

believing a tool that says “it’s passed on”, with no trace of the call;
accepting a command that looks for a key “to sign in”;
letting two agents write in the same folder at the same time;
installing a plugin “that links everything” without checking where it comes from (module 7, lesson 6).

Keep it and pick it up later

Keep the map of exchanges in your journal, next to “My extensions”.
With each new tool, add its line.

Going further

optional

Codex’s documentation recommends creating Git checkpoints before and after a task, so you can go back: it is the method from module 5, lesson 6.

New words

4 words explained

Word	What it means
Integration	A documented, technical link between two tools
Copy-paste	Passing the text from one tool to the other yourself
Shared files	Two tools that work, one after the other, in the same folder
Access key (API key)	A secret code that lets software use a service in your name, and at your expense