MelıAI Mentoring
Use CasesSkills

Course material

  • CoursesPractice AI skills through focused lessons.
  • Learning pathsTake courses in a useful order.
  • Skill treeThe skill map behind every course.

Resources

  • AI TermsPlain definitions and quick practice.
  • GuidesDetailed help for useful AI work.
  • WalkthroughsCurrent setup steps for useful tasks.
  • Self-guidedMeli AI AcademyFree tierPublished courses and guides at your own pace.Create a free account
  • GroupAI Mastermind$97/monthLearn together in live group sessions.Explore AI Mastermind
  • One-on-oneIndividual coachingLimited availabilityIndividual AI coaching and consulting.Request availability
About Me

Your bookmarks

Loading your bookmarks

Latest updates

New courses, skills, book chapters, and site improvements will appear here.

Create free accountSign upDashboardDashboard
Use CasesSkillsAbout Me

Learn

Course material

  • CoursesPractice AI skills through focused lessons.
  • Learning pathsTake courses in a useful order.
  • Skill treeThe skill map behind every course.

Resources

  • AI TermsPlain definitions and quick practice.
  • GuidesDetailed help for useful AI work.
  • WalkthroughsCurrent setup steps for useful tasks.

Offerings

  • Self-guidedMeli AI AcademyFree tierPublished courses and guides at your own pace.Create a free account
  • GroupAI Mastermind$97/monthLearn together in live group sessions.Explore AI Mastermind
  • One-on-oneIndividual coachingLimited availabilityIndividual AI coaching and consulting.Request availability
Sign in
More links below

AI coaching to improve your impact and improve any area of your life.

MelıAI Mentoring

I never sell your personal data.

Learn

  • Courses
  • Learning paths
  • Skill tree
  • Skills

Resources

  • Use Cases
  • AI Terms
  • Guides
  • Walkthroughs
  • Security audit

Work with me

  • Meli AI AcademyFree tier
  • AI Mastermind$97 a month
  • Individual coachingLimited availability

About

  • About Me
  • Contact
  • Member sign in
Privacy PolicyAcademy DataCoaching PrivacyTerms of Use

© 2026 Meli AI Mentoring. All rights reserved.

Walkthroughs/Any model/Custom app connections

Connect your agent to an app with no connector or plugin

Let your agent help you read or update another app, even when no ready-made connector or plugin does your task.

12 minutesIntermediateAny model
Outcome

A connection that has done one small task, with access you chose and a result you checked.

Follow the steps

Member starter

Members only

Start with your agent

Copy the directions into your AI app. Your agent can help with the setup; the steps below explain what to do and what to check.

Sign in to copy

New here? View membership

How it works

Suppose you want your agent to list your upcoming events or draft a new one in another app. The usual route is a ready-made link your AI app offers: Claude calls it a connector, and ChatGPT and Codex call it a plugin. When none exists for your app, or none can do your particular job, your agent may still be able to do the task.

An API lets other software ask an app for data or actions, such as listing events. An MCP connection uses the Model Context Protocol to give an AI application access to information or actions exposed by an MCP server. It may be supplied by an existing integration or configured separately. Computer use lets an agent operate on-screen buttons and fields the way you would, including on a web page in a browser. Your agent can check which route supports your task and help set it up. You choose the account and what it may access, then check the result of one small test before relying on it.

  1. 01Find a way to do your task
  2. 02Choose what the connection can access
  3. 03Ask your agent to prepare the action
  4. 04Check a small result in the app

Connect the app, then try one small task

  1. 01

    Find a way to do your task

    You have a particular job in mind, in a particular app. Before building anything, check that some route can do that exact job, because an app may let software read an event but not publish one.

    Tell your agent the app and action, such as “list my upcoming Eventbrite events.” Ask it to compare the routes that support it and explain which needs the least setup.

    Ask your agent to find a connection

    Members can copy this prompt. Sign in or view membership.

    Check: The agent names a supported option, links to its official instructions, and explains any account requirements.

    The Eventbrite API reference with the API title and OAuth2 authentication highlighted.The Eventbrite API reference with the API title and OAuth2 authentication highlighted.
    An example of the API route: Eventbrite's developer documentation lists what other software can ask for.
  2. 02

    Choose what the connection can access

    Before the connection works, you choose which account it uses and what it may do there, such as read only. Often you grant this by signing in to the app. Some routes instead need an API key or token, a secret code that works like a password for software. Anyone holding it may be able to use that access, so it never goes in the chat.

    Ask your agent to guide the sign-in and explain each permission. If the route needs a key or token, enter it yourself where your agent recommends, outside the chat.

    Get help with sign-in and permissions

    Members can copy this prompt. Sign in or view membership.

    Check: You know which account is connected and whether it can read, create, or change items.

    For a work account, follow your organization’s rules. If a setting you need is locked, ask your administrator.

    The Eventbrite Help Center article with Log in, Generate an API Key, and Account Settings highlighted.The Eventbrite Help Center article with Log in, Generate an API Key, and Account Settings highlighted.
    Eventbrite's help page for creating an API key. Eventbrite reviews each request and emails you when the key is approved. Many connections use a normal sign-in instead; follow the current instructions for yours.
  3. 03

    Ask your agent to prepare the action

    With access set up, the agent needs to know exactly what to ask the app for. A connector, plugin, or MCP connection may already include that action. On the API route, the agent may write a small script, a saved set of computer instructions, that sends the request.

    Describe one thing to read or change and the result you expect. For example: “List my upcoming events and save their names and dates in my project notes.” Ask the agent to prepare that action through the route you chose.

    Prepare the app action with your agent

    Members can copy this prompt. Sign in or view membership.

    Check: The agent explains what it will read or change and gives instructions for running the action again.

    An unbranded agent request, plan, and connection script explainer.An unbranded agent request, plan, and connection script explainer.
    The agent's plan and the script it writes for the API route.
  4. 04

    Check a small result in the app

    A successful sign-in only shows that the connection can get in. To learn whether it does your job correctly, have it read an item you know well and compare the details with the app. If your task creates or changes something, have it make a draft instead and check the draft in the app.

    Ask your agent to state the expected result and run the agreed test. Open the app yourself and compare the item with what the agent reported.

    Test the connection with your agent

    Members can copy this prompt. Sign in or view membership.

    Check: The returned information or draft matches what you see in the app. If it does not, show your agent both versions and ask it to find the cause before you continue.

    A dry-run test request creating a draft that is reviewed before publishing.A dry-run test request creating a draft that is reviewed before publishing.

Understanding

Each connection can do only what its permissions allow

A connection might be allowed to read your events but not change them. Some actions also depend on your account plan or an administrator. Ask your agent to explain what the connection can do, then grant only the access your task needs. A successful test shows only that the action you tested works with the access you granted.

An agent request, API credential, connection code, and platform API shown as one flow.An agent request, API credential, connection code, and platform API shown as one flow.

If you get stuck

There is no connection that supports my task.

Ask your agent to recheck official sources for connectors or plugins, MCP connections, the app's API, and any browser or computer use the app's terms allow. If none supports the action, do that step by hand rather than working around the app’s restrictions.

The app says I do not have permission.

The app knows who is asking but will not allow this action. Share the error with your agent, leaving out the key or token, and ask it to check the account, granted permissions, plan, and administrator settings against the official docs.

The connection fails with an authentication error.

The app does not accept the sign-in, key, or token. Share only the error, and ask your agent to check the secret's stored name and location, its expiration, and the sign-in method, without printing it.

Sources and further reading

Official sources and check dates

The dates below show when each official source was checked against the steps on this page.

MCP architecture overview

Defines the host, client and server, and the tools, resources and prompts that servers can expose through the protocol.Model Context Protocol technical documentation•Checked November 6, 2026

Model Context Protocol

MCP connections expose supported tools to Codex and may use different sign-in methods, including OAuth, bearer tokens, and environment variables.OpenAI•Checked September 22, 2026

Computer Use

Supported computer-use setups let Codex interact with on-screen app controls.OpenAI•Checked September 22, 2026

Connectors overview

Documents that connectors link Claude to outside services through the Model Context Protocol (MCP), from prebuilt connectors to custom remote MCP servers.Claude.ai Documentation•Checked September 22, 2026

Instagram Platform: Content Publishing

Documents the content publishing flow, required permissions, and rate limits for the Instagram API.Meta for Developers•Checked September 22, 2026

Eventbrite Platform introduction

Introduces Eventbrite's supported developer platform and links the current API documentation.Eventbrite Platform•Checked July 10, 2026

Eventbrite API authentication

Documents the current authentication flow and token requirements for Eventbrite API requests.Eventbrite Platform•Checked July 10, 2026

Notion API connections

Documents that Notion uses internal connection tokens with granular capabilities, such as read, update, and insert content, showing that app access goes beyond one simple API key.Notion Help Center•Checked September 22, 2026

Find your API token

Shows where to find Todoist's personal API token, a platform-specific credential type distinct from a generic API key.Todoist Help Center•Checked September 22, 2026

Get started with connectors

Documents the current connector flow: choose a connector, authenticate through the provider, review access, and use the connected service in a conversation.Claude.ai Documentation•Checked September 22, 2026

Get started with custom connectors using remote MCP

Documents custom connectors for services without a ready-made connector, which Anthropic has not verified, and how availability differs by plan.Claude Help Center•Checked September 22, 2026

Plugins

Documents that ChatGPT and Codex plugins bundle skills and MCP servers, and that a plugin can prompt you to connect an external service before its tools become available.OpenAI•Checked September 22, 2026

Let Claude use your computer in Cowork

Documents Claude computer use: Claude sees the screen, clicks, types, and opens apps after asking permission for each app.Claude Help Center•Checked September 22, 2026