All roads lead to Customer.io: Every way to engage with AI 

Say you're a lifecycle marketer with a Tuesday-morning idea: pull everyone who stopped opening emails 60 days ago and send them a campaign to re-engage. A developer on your team has a different Tuesday, wiring the iOS SDK into a new app. Your Head of Growth wants last week's broadcast numbers before a 10 a.m. meeting. All three need Customer.io, and each of them reaches it a different way.

Those different roads exist because AI has become part of how marketing teams get work done. In Customer.io's 2026 State of Customer Messaging survey of 750+ marketers, product managers, and growth leaders, 73% of marketers reported meaningful AI impact on their messaging strategy. With that many teams working alongside AI, the useful question is which road fits which job. You can reach Customer.io through the Agent in the app, a Slack agent, MCP in tools like Claude and ChatGPT, and a CLI built for coding agents.

Picture a yellow brick road with a few on-ramps, all ending at the same Emerald City: a behavior-triggered message tied to a business outcome, like completed onboarding or a paid plan, that reaches someone at a moment that matters to them. Keeping that message performing takes experimentation, a full-time job that everyone runs part-time between deadlines, and no dashboard shows the test you didn’t run.

Customer.io has helped teams optimize toward business outcomes since day one, and each road on this map makes that loop faster. The Agent already knows your business, your data, and your history, so you can ask it to optimize an automation or pull up how an A/B test performed without starting from zero. LLM actions let a journey personalize a message or route someone down a different path in the moment.

TLDR:

• Your Agent in Customer.io is the best fit for marketers who build and analyze inside the workspace. It can do nearly everything you can do in Customer.io, from creating segments and automations to answering performance questions, and it remembers your preferences between conversations.
• The Slack agent is the same Agent, reached from a thread. It suits quick questions and reporting where your team already talks, runs with each person's own Customer.io permissions, and creates drafts by default.
• The MCP server connects Claude, ChatGPT, Cursor, and other AI tools to the full Journeys UI API and CDP Data Pipelines API, and it works whether or not in-app AI is turned on.
• The CLI gives coding agents like Claude Code structured, self-describing access to your workspaces from the terminal, including SDK setup.

There's no place like home, so start with the surface closest to where you already work: the Agent if you work in Customer.io, the Slack agent if your team works in Slack, MCP if you work in a chat-based AI tool, and the CLI if you work from a terminal.

Your AI agent in Customer.io is a coworker that lives in the app, and it can do nearly anything you can do there: build segments, set up automations and one-time sends, draft Design Studio emails, and dig into your metrics. Click Agent in the sidebar or Ask Agent in the bottom right corner and start typing.

Specific prompts get the best results, and a stated goal helps the Agent suggest a complete approach. These come straight from the docs and are worth borrowing:

• "Create a segment of people who signed up in the last 30 days but haven't completed onboarding."
• "What's the click-to-open rate for my Onboarding Welcome automation this month?"
• "I want to re-engage people who haven't opened an email in 60 days."

Because the Agent reads your metrics and reporting data, it can trace conversions back to specific messages and answer performance questions the metrics pages leave open. It also remembers how you work. Tell it to always use sentence case in your subject lines, or to include bounce rate and spam complaint rate whenever you ask about deliverability, and it applies those preferences in later conversations. Upload a brand style guide, a CSV of customer data, or a screenshot of an automation from another platform, and the Agent will follow the guide, analyze the data, or rebuild the automation.

Routines put the Agent on a schedule. A deliverability watchdog or a segment hygiene check lands in your inbox on its own, weekly on Essentials plans and daily on Premium and Enterprise plans. Routines run in safe mode, so they read and report without changing anything.

The in-app Agent is the one to reach for when you want to watch the work happen. The curtain stays open: you see it work in real time and review every draft in the same place you'd build it yourself. It works inside Customer.io, so it can't query your external databases or warehouses or browse the web, but it can help you set up the integrations that bring that data in.

The Slack agent is the same Agent, reached with an @Customer.io mention or a direct message. A head of growth in the lifecycle channel wonders how last week's broadcast compared to the one before it. A quick mention gets the answer in a thread, and the Agent can post the findings to a channel or send a teammate the numbers. Each thread is its own conversation, so reply in the thread to keep your context or mention the Agent somewhere new to start fresh.

An account admin installs the Slack app once from Settings > Slack Agent, and anyone with a Customer.io login can use it from there. The Agent runs as you, with your permissions, and every action lands in your audit logs as Slack Agent (on your behalf). It creates drafts by default. When something needs to change in live data, it posts an Approval needed card that names the change, the permissions it wants, and the workspace, and your approval covers that one change for about five minutes.

Admins can also switch on Read-only in Slack, which turns the Slack agent into a reporting assistant that answers questions about automations, segments, and people without creating or editing anything. A few other details set it apart from the in-app Agent:

• It can't read sensitive profile attributes
• You get messages at milestones with a final answer at the end
• Slack conversations stay in Slack and don't appear in your Customer.io conversation history

MCP makes more sense when your work starts in an AI tool you already use and needs to reach into Customer.io partway through. Customer.io MCP connects your account to tools that support the Model Context Protocol, including Claude, ChatGPT, and Cursor, and gives them the full Journeys UI API and CDP Data Pipelines API. From a chat window, you can read workspace data, create and manage automations and one-time sends, and work with segments.

A few prompts from the docs show the range:

• "Show me the delivery metrics for my onboarding automation."
• "Draft a welcome email in Design Studio."
• "My users aren't receiving push notifications on Android. Can you help me troubleshoot?"

Setup depends on your tool. Claude Code and Cowork users install the Customer.io plugin for Claude Code, ChatGPT and Codex users install the Customer.io plugin, and you can also set up the Claude chat connector or the ChatGPT connection directly. An account admin turns on the Customer.io MCP toggle once, and each person then signs in with their own Customer.io login.

Permissions come in layers. A new connection starts with read access only. Your AI tool asks for write, write:live, configure, or read:sensitive when it needs them, and you approve or deny each one. You can give your AI tool less access than you have yourself, and never more. Reads, writes, and deletes also run through separate tools, so a client with tool-level permissions, like Claude's Connectors UI, can approve reads automatically and check with you before every write.

MCP works independently of in-app AI, so teams that keep the Agent and other in-app features switched off can still use Customer.io through external tools their organization approves. Since MCP tools can return information about people in your workspace, connect only AI tools your organization has approved.

The Customer.io CLI fits developers and data partners who work with coding agents like Claude Code, and anyone building scripts and automation around Customer.io. For the Scarecrow on your team, the one who came to Oz for a brain, this is the road. It was designed for LLMs first: every command returns structured JSON, every endpoint describes itself through cio schema, and a single cio api command covers the Journeys UI API and the CDP Data Pipelines API.

Getting started takes three steps. Install the CLI with npm install -g @customerio/cli, run cio skills install to teach your coding assistant how to drive it, and open Claude Code, Cursor, or a similar tool in your project. Then ask in plain language:

• "Set up Customer.io for my React Native app."
• "Find people who haven't opened an email in 30 days and create a win-back segment."

The CLI's sub-skills cover account onboarding and SDK integration, so a single entry point handles signup, domain setup, a first email, and wiring SDKs into your code.

Authentication uses service account tokens. A user token takes its permissions from the person it belongs to and can be locked to read-only. A system token carries a role you assign, which suits CI/CD pipelines and scripts. If you're choosing between the CLI and MCP, the docs offer a clean rule: use the CLI when you're working in a terminal or with an agent-driven coding tool, and use MCP when you want a quick connection from a chat-based tool without installing anything locally.

Beyond the four surfaces, a few extra gifts from the Wizard's bag are built into everyday features:

• LLM actions add steps to your journeys that send data through a large language model and store the result as a journey attribute, which powers personalized recommendations, sentiment analysis, and translation. They use AI credits.
• Email content analysis reviews your email and suggests better ways to deliver your message.
• Recommended send time picks a send time based on your message content and your audience's time zones.
• In-app survey analysis distills survey responses into insights you can act on.
• AI translation drafts a new language variant from your default content.
• Data descriptions generate plain-language descriptions of attributes and events in your Data Index.
• Workspace search (⌘K or Ctrl K) uses AI-enhanced ranking to find automations, segments, and templates.

Clear names and descriptions on your data pave the road for every AI surface, because the AI reads them the way a new teammate would. An attribute called cname could mean a company name, a customer name, or a canonical domain name, and a one-line description in your Data Index settles the question for the Agent, for MCP, and for your colleagues. You can generate that description with AI right from the Data Index.

Four habits pay off across every surface:

• Set your business context. Customer.io builds a starting profile from your sending domain, and you can edit your business and audience descriptions, tone and voice, and key links under Workspace settings > Business context. Each workspace has its own profile.
• Name things consistently. Pick snake_case or camelCase for attributes and events, and AI suggestions follow the same format.
• Tag your assets. Tags organize automations and segments, and tag descriptions give the AI more to work with.
• Describe your attributes and events. Descriptions are the fastest way to take guesswork out of your prompts.

Customer.io's AI features use the names and descriptions of your data, which means the key and not the value. The AI knows you have a first_name attribute and never sees that the person's name is Alex.

Each surface shortens the loop between an idea and a live, behavior-triggered message, and every pass through that loop moves you toward optimized outcomes.

The marketers in Customer.io's latest report point to the same destination. Segmentation (31%) and behavioral triggers (29%) led the list of tactics that drive ROI, with everything else below 10%, and 66% of marketers track conversion rates as their primary success metric. A faster loop gives you more chances to refine the segment, the trigger, and the message against those conversions.

Explore AI in Customer.io or book a demo to see which road gets your team to the Emerald City fastest.