Set up your agent:
one page, every agent.

ChatGPT

01Create your account

Open chatgpt.com in your browser and sign up. Use an email address you actually check: login codes and important notices go there.

02Get the app

Install the ChatGPT app from the App Store (iPhone) or Google Play (Android) and log in with the same account. The app adds voice, image, and camera input on top of chat.

03Connect your first tool

Open Settings and look for Apps / Connectors. Start with a prebuilt one: Gmail, Google Drive, or Calendar. Grant access once and ChatGPT can read from it when you ask.

For your own tools (APIs, MCP servers), you need a paid plan: Plus, Pro, or Business. On a paid plan, turn on Developer Mode in Settings, then add a remote MCP server by its URL. The server's tools appear in your chats.

04Run the test

Ask one question only the connected tool can answer. Then read guide 004 to connect your first API in 10 minutes: the same three moves (get access, connect it, test it) work here.

Next: connect your first API →

Claude

01Create your account

Open claude.ai in your browser and sign up. Use an email address you actually check: login codes and important notices go there.

02Get the app

Install the Claude app from the App Store (iPhone) or Google Play (Android) and log in with the same account.

03Connect your first tool

Open Settings → Connectors. Start with a built-in one: Gmail, Calendar, Google Drive, Slack, or Notion. Grant access once and Claude can read from it when you ask.

For your own tools, add a custom MCP server: in Connectors, add a remote MCP server by its URL. For local servers and coding work, use Claude Code from the terminal: claude mcp add walks you through it.

04Run the test

Ask one question only the connected tool can answer. Then read guide 004 to connect your first API in 10 minutes: the same three moves (get access, connect it, test it) work here.

Next: connect your first API →

Codex

01Create your account

Sign up at chatgpt.com if you have not already. Codex rides on that account, so there is nothing new to register. On a free plan you may see limited trial access; the full agent unlocks on Plus or higher.

02Get the app

Install the Codex app for your computer, or install the CLI with npm i -g @openai/codex if you prefer the terminal. Sign in with your ChatGPT account when it asks. The app, the CLI, and the editor extension all read the same settings.

03Connect your first tool

Codex reads MCP servers from ~/.codex/config.toml, and note the format: TOML, not the JSON most other agents use. The easy path is to run codex mcp add and let the CLI write the entry. By hand, a server is a block that starts with [mcp_servers.name] followed by its command or URL. Restart Codex after editing the file.

04Run the test

Give it one small job that needs the connected tool, in a project you can afford to experiment on. Then read guide 004 to connect your first API: the same three moves work here.

Next: connect your first API →

Copilot

01Create your account

Sign up at github.com. Copilot activates against that account, and the free tier is enough to follow this whole guide.

02Get the app

Install VS Code, sign in with your GitHub account, and add the GitHub Copilot and Copilot Chat extensions from the marketplace. Open the Chat view and pick Agent in the mode dropdown.

03Connect your first tool

Add an MCP server in a file named .vscode/mcp.json inside your project. Heads up: the top-level key in this file is "servers", not the "mcpServers" most other agents use. Or run the Command Palette action MCP: Add Server and let VS Code draft it. Connected tools appear in the agent's tool picker.

04Run the test

Ask the agent one question only the connected tool can answer. Then read guide 004 to connect your first API: the same three moves work here.

Next: connect your first API →

Cursor

01Create your account

Sign up at cursor.com, or create the account inside the app when it first opens. Use an email you actually check.

02Get the app

Download Cursor for Mac, Windows, or Linux and sign in. Open one of your project folders so the agent has something real to work on.

03Connect your first tool

Open Settings, find MCP, and choose Add new global MCP server. Or edit ~/.cursor/mcp.json directly and add your server under "mcpServers". For one project only, put the same file at .cursor/mcp.json in the repo. A green toggle next to the server means its tools are live.

04Run the test

Ask the agent one question only the connected tool can answer. Then read guide 004 to connect your first API: the same three moves work here.

Next: connect your first API →

Devin Desktop

01Create your account

Sign up in the app or at devin.ai with the account you will code with.

02Get the app

Download Devin Desktop for your computer and sign in. Open a project folder and find the Cascade panel: that chat is your agent.

03Connect your first tool

In the Cascade panel, open the ... menu in the MCPs section and choose Open MCP config file. Current versions keep it at ~/.config/devin/mcp_config.json (older Windsurf installs used ~/.codeium/windsurf/mcp_config.json). Add servers under "mcpServers". One gotcha: remote servers take the key "serverUrl" here, not "url".

04Run the test

Ask Cascade one question only the connected tool can answer. Then read guide 004 to connect your first API: the same three moves work here.

Next: connect your first API →

Gemini

01Create your account

A Google account is the whole signup. Open gemini.google.com and sign in with it.

02Get the app

Install the Gemini app on your phone for chat and voice. For tools, install the CLI on your computer with npm i -g @google/gemini-cli and sign in with the same Google account; the CLI includes a free tier.

03Connect your first tool

In the app, open Settings and turn on the connected apps you want (Gmail, Drive, Maps). For your own MCP servers, open ~/.gemini/settings.json and add them under "mcpServers", then start the CLI and run /mcp to confirm they loaded.

04Run the test

Ask the CLI one question only your tool can answer. Then read guide 004 to connect your first API: the same three moves work here.

Next: connect your first API →

Genspark

01Create your account

Open genspark.ai in your browser and sign up. Use an email address you actually check: login codes and important notices go there.

02Pick your first job

Genspark has a lot of surfaces: chat, research, slides, sheets, images, video, browser. Do not try them all at once. Pick one real job: a research brief, a slide deck, or a comparison sheet. One finished artifact teaches you the workflow faster than five half-started ones.

03Connect your first tool

Open the Tools section / MCP Store. Start with a service you already use: Gmail, Calendar, Notion, Slack, Outlook, X, or Reddit. Grant access once and Genspark can read from it when you ask. The MCP Store lists hundreds of tools; add the ones that solve problems you actually have.

04Run the test

Ask one question only the connected tool can answer. Then read guide 004 to connect your first API in 10 minutes: the same three moves (get access, connect it, test it) work here.

Next: connect your first API →

Grok

01Create your account

Open grok.com in your browser and sign up, or open the Grok app and sign in. You can also reach Grok inside the X app.

02Get the app

Install the Grok app from the App Store (iPhone) or Google Play (Android) and log in with the same account.

03Connect your first tool

On grok.com, open Connectors, then New Connector, then Custom. Add a remote MCP server by its URL. The server's tools appear in your chats.

Want the teammate version? Grok Bot is a separate desktop app with its own plugins and skills, under Settings → Plugins. Start with grok.com first; add Grok Bot when you want an always-on teammate.

04Run the test

Ask one question only the connected tool can answer. Then read guide 004 to connect your first API in 10 minutes: the same three moves (get access, connect it, test it) work here.

Next: connect your first API →

Instinct

01Get an invite

Two doors. Door one: join the waitlist at app.instinct.com and wait for your invite. Door two: get an invite link from someone who already has Instinct. Each member gets 5 invites to share. If you know a member, door two is faster.

02Sign up with your phone

Open your invite link and sign in with your phone number plus a text code. You must be 18 or older. There is no app to install: Instinct lives in your messages (iMessage, WhatsApp, SMS) and on calls.

03Connect your accounts

Connect your email, calendar, and messaging apps. This is what turns Instinct from a chatbot into an agent: with your accounts connected, it can book, buy, follow up, and call on your behalf instead of just answering questions.

04Hand it one real job

Give it something that needs action, not just an answer: book a table, follow up on an email thread, research and buy a product. Instinct's agents coordinate with each other through a trusted network (Instinct-to-Instinct only), including group chats where your friends do not need accounts.

Next: connect your first API →

Manus

01Create your account

Open manus.im in your browser and sign up. Use an email address you actually check: login codes and important notices go there.

02Give it one real task

Do not start with a test question. Hand it something from your actual life: research a purchase, plan a trip, organize a project. Manus works best when it has a job with a finish line, not a chat to fill.

03Connect your first tool

Open Settings → Integrations. Start with an OAuth Connector (a service you already use), or add a custom MCP server: Custom MCP Servers takes a hosted HTTPS URL plus a Bearer token. The server's tools appear in your Manus chats. Skills (imported from GitHub or uploaded) extend what it can do.

04Run the test

Ask one question only the connected tool can answer. Then read guide 004 to connect your first API in 10 minutes: the same three moves (get access, connect it, test it) work here.

Next: connect your first API →

Muse

01Create your account

Open muse.ai in your browser and sign up. Use an email address you actually check: login codes and important notices go there.

02Get the app on your phone

Search for Muse in the App Store (iPhone) or Google Play (Android) and install the official app. Log in with the same account you just created.

Prefer the browser? That works too. The point is having your agent one tap away instead of buried in a tab you never open.

03Connect WhatsApp (optional, recommended)

In Settings, find WhatsApp linking and follow the prompts to connect your number. This puts your agent inside the app you already check fifty times a day.

04Give it one real task

Do not start with "what can you do." Start with something from your actual life.

Next: connect your first API →

OpenClaw

01Install Node.js 22 or newer

OpenClaw runs on Node.js, version 22 or higher. Check what you have:

node --version

No Node yet, or the version is too old? Install the LTS release from nodejs.org, then close your terminal completely and open a fresh one.

02Install OpenClaw

On macOS or Linux, run the official installer:

curl -fsSL https://openclaw.ai/install.sh | bash

On Windows, open PowerShell and run:

iwr -useb https://openclaw.ai/install.ps1 | iex

Prefer npm? This works too:

npm install -g openclaw@latest

03Run the onboarding wizard

openclaw onboard --install-daemon

The wizard walks you through three decisions in plain words:

1. Which AI model powers your agent. Pick Anthropic, OpenAI, OpenRouter, or a local model via Ollama. The hosted options need an API key; the wizard asks for it and stores it securely.
2. How the gateway runs. Accept the default: the wizard installs it as a background service so your agent keeps working when the terminal is closed.
3. Which chat channels to connect. WhatsApp, Telegram, Discord. You can add these later too.

04Check that it is alive

openclaw gateway status

openclaw health

05Open the dashboard and say hello

openclaw dashboard

Or open http://127.0.0.1:18789/ directly in your browser. This is your agent's control room: chat, integrations, and settings in one place.

06Connect your phone (optional)

openclaw channels login

Follow the prompts to link WhatsApp or Telegram. Message your agent from your phone and get a reply.

openclaw gateway --port 18789

sudo loginctl enable-linger $USER

Next: connect your first API →

Poke

01Start Poke

Open poke.com in your browser and follow the prompts to start texting Poke. It runs inside iMessage, SMS, or WhatsApp: the agent lives in your messages, not in a separate app you have to remember to open.

02Set your price

Poke asks you to name your monthly price. Reported prices run roughly $2 to $29 a month depending on the haggle. Offer what the service is worth to you: lower is fine, and Poke will counter if it wants more. This is the one agent where the price is a conversation, not a price list.

03Connect your first integration

Connect a service you already use: Gmail, Calendar, Notion, Linear, or Oura. Grant access once and Poke can read from it when you ask. Poke also supports Recipes (saved automations) and MCP for custom tools.

04Run the test

Text Poke one question only the connected integration can answer. Then read guide 004 to connect your first API in 10 minutes: the same three moves (get access, connect it, test it) work here.

Next: connect your first API →