AI connectors

What do you want to do?

These are different mechanisms. A Skill provides instructions and scripts for a task; MCP allows an application to invoke P2LRN tools. None replaces the Player or the visual Studio editor. Choose your platform first, then follow the installation steps for that environment.

Create and download a course with a Skill

The “P2LRN – Create and download” Skill can generate and verify new packages from the Core 0.4.0 subset. Downloadable versions are available for ChatGPT, Claude, Gemini CLI and Grok Build. Creating an actual course package requires an environment capable of running the included Python packager; downloading a Skill does not install it or publish it in any catalog.

1. In your AI app, open the Skills feature and check that you have a P2LRN-compatible version installed.
2. Specify the title, language, structure, quizzes and games. Ask for an actual file and not just a content outline.
3. Verify that the packager and validator have been run. Download the .p2lrn and try it on Player or Studio.

Create a P2LRN introductory science course in English:
3 units, 6 lessons, 10 questions, and 2 games.
Run the P2LRN packager and validate the result.
Give me the downloadable .p2lrn file.

If your application cannot run the packager, it may help plan the course, but it must not claim to have generated a valid package. You can also create and edit the course visually in Studio.

ChatGPT

Create courses

With the “P2LRN – Create and Download” Skill installed and enabled, describe the course you want in the conversation and request the verified file. Skill availability depends on the settings and permissions of your account or workspace. Official Skills Help.

Work with local courses

ChatGPT in the browser cannot open directly localhost. To use the private connector from ChatGPT, check the option safe tunnel when it is enabled in your account. There are also MCP servers accessible via HTTPS with authentication. The connection is not installed automatically and is not equivalent to an approved public plugin.

Test: ask “List my authorized P2LRN courses and validate one I choose.” You must see a real invocation of p2lrn_list_courses and p2lrn_validate_course, not a fabricated response.

Claude

Create courses

Claude supports Custom Skills in environments that have that feature enabled. Review the Skill before importing it and confirm that its scripts can run and deliver files. The existing P2LRN Skill in ChatGPT is not automatically certified for Claude. Official Claude Skills Guide.

Work with local courses

In Claude Desktop, you can use the local connector: select Claude in the wizard, authorize a folder and start reading. In Claude web, external connections depend on the plan and the custom connectors supported.

Test: ask “Read the outline of an authorized course and suggest improvements without modifying it.” Confirm that the client really accesses only the permitted course and respects read-only permissions.

Gemini

Gemini in the browser

Gemini web has customization functions such as Gems and, depending on account, region and deployment, Skills. These options can prepare the course content, but to download a .p2lrn the session must be able to run and verify a supported packager. The P2LRN Skill in Gemini web has not been certified here.

Gemini CLI

Gemini CLI supports Agent Skills using files SKILL.md. You can check them from the console with gemini skills list and link an authorized Skill folder. To consult courses on your computer you can also configure the P2LRN connector by selecting Gemini CLI in the setup assistant. These are separate options.

.\integration\mcp\Conectar-P2LRN.cmd --root C:\MyCourses --clients gemini --install --verify

C:\MyCourses is an example path that must exist. Run this command from the extracted MCP kit folder. Test: ask Gemini CLI to list and validate an authorized course and confirm actual MCP tool calls.

Grok

Create content

A Skill for Grok Build is available. Its packager requires Python to create and validate files .p2lrn. This is not equivalent to a certified Skill for Grok web. If your environment does not run the packager, complete the course at Studio.

Connect courses

When your account supports custom MCP connectors, see the configuration options. Grok Connectors and register an authenticated, securely accessible server. The P2LRN kit allows you to generate connection descriptors; it does not install the connector in your account or prove that it has been approved.

Test: ask "List my authorized P2LRN courses and validate one without editing it." Check that Grok really uses the tools and respects the allowed folder.

Install the connector on your computer

The connector allows a compatible application to work with courses saved in a folder on your PC. It works with compatible clients such as Claude Desktop, Gemini CLI, Cursor, Windsurf and VS Code. You don't need to have a website or buy a domain.

1. Install Node.js 22 or later and unzip the ZIP to a folder on your computer.
2. Run integration\mcp\Conectar-P2LRN.cmd. Choose the installed AI application you want to use.
3. Select a course folder and start with read permissions (P2LRN_MCP_WRITE=0). Authorize writing only when you need it.
4. Restart the application if prompted and test that it can list and validate one of your courses.

cd C:\P2LRN-MCP
.\integration\mcp\Conectar-P2LRN.cmd

C:\P2LRN-MCP is an example installation path. The wizard asks permission before configuring supported clients. The ZIP is neither a Skill nor a public plugin.

Connect from a web service

When you use AI from its website, the service cannot directly read the files on your PC. You need an authorized private connection or a server accessible via HTTPS. Only prepare this connection if you are going to work with existing courses from a remote service.

$env:P2LRN_MCP_ROOT = 'C:\CursosP2LRN'
$env:P2LRN_MCP_WRITE = '0'
$env:P2LRN_MCP_TOKEN = '<token-aleatorio-seguro>'
$env:P2LRN_MCP_PUBLIC_ORIGIN = 'https://your-endpoint.example'
node .\integration\mcp\server.mjs --http

https://your-endpoint.example It is a bookmark and not a published service. The local server listens on 127.0.0.1:37871/mcp; for a public URL you need to configure and secure the proxy. A private ChatGPT tunnel requires its own configuration and testing.

Common checks and problems

• The AI does not deliver a file: check that the Skill runs a packer and returns a valid .p2lrn. A text or JSON response is not enough.
• Your courses do not appear: check the authorized folder, the MCP client and that the application has loaded the configuration.
• The website does not connect to your PC: check HTTPS tunnel or endpoint availability and your account permissions; localhost is not accessible from an external website.
• The application cannot edit: check permissions. Reading is the initial mode; drafts require explicit authorization.
• Minimum test: run p2lrn_list_courses and p2lrn_validate_course, checks for rejection of unauthorized paths and confirms that the original file does not change.

If you want to create or edit a course manually, you can use Studio. To play it and check its activities, use Player.