Table of Contents

How to build, test and publish an agent with the Landbot MCP or the Landbot Plugin

Desirée M Updated by Desirée M

Describe the conversation you want in everyday words, and your AI assistant or AI coding tool builds the Landbot agent, lets you test it and publishes it when you approve. You don't need to know block names or how the builder works.

How to build an agent with the Landbot MCP or the Landbot Plugin

1️⃣ Open a new chat in your AI assistant (with the Landbot MCP), or in Claude Code or the Claude desktop app (with the Landbot Plugin).

2️⃣ Describe the agent: who it talks to, what it asks, in what order, and what happens after each answer. For example:

Build a lead qualification agent for our website. Greet visitors, ask for their name, email and company size. Tell companies with more than 50 people that someone from sales will contact them, and thank everyone else. Save each lead to a Google Sheet.

3️⃣ Sign in to Landbot if your browser asks you to.

4️⃣ Wait while the agent is built. When it's ready, you get a plain-words summary of the conversation.

Claude's summary of a lead qualification agent: name, email, company size buttons and two endings

5️⃣ Choose whether to test the agent first or publish it. Visitors only see the agent after you approve publishing.

How does the Landbot Plugin build a new agent?

The Landbot Plugin builds a new agent in a fixed order, so nothing goes live until you approve:

1️⃣ The Landbot Plugin builds the conversation.

2️⃣ The Landbot Plugin switches the agent to the current web chat and applies the look you described. The agent isn't published yet, so no visitor sees it.

3️⃣ The Landbot Plugin asks once for your approval and publishes the agent.

4️⃣ In the Claude desktop app, the Landbot Plugin tests every path of the conversation in a browser panel next to the chat.

5️⃣ The Landbot Plugin gives you the builder link, the share link and a plain-words description of the conversation.

For an agent you already have, the Landbot Plugin names the agent and asks for your approval before changing it.
Not sure where to start? With the Landbot Plugin, say "just show me." The Landbot Plugin builds a sample lead qualification agent and publishes it after one approval, so you can try it right away.

Which Landbot blocks can't the Landbot MCP and the Landbot Plugin build yet?

Not every Landbot block can be created with the Landbot MCP or the Landbot Plugin yet. If you ask for an unsupported block, they say the block isn't supported instead of using a different one, and you can add it in the Landbot builder.

● Bricks: the Landbot MCP and the Landbot Plugin can't create Bricks. An agent's existing Bricks are kept as they are.

● Yes or no questions and ratings: on the current Landbot web chat, the Ask for a yes or a no block and the Ask for a rating block don't appear to visitors, so the Landbot MCP and the Landbot Plugin build those questions with the Buttons block instead.

● Other blocks: Landbot adds support for more blocks over time. To use a block that isn't in the list below, add it in the Landbot builder.

Which Landbot blocks can the Landbot MCP and the Landbot Plugin build?

As of 13 October 2026, the Landbot MCP and the Landbot Plugin can build these Landbot blocks:

● Questions: Ask a question, Ask for a name, Ask for an email, Ask for a phone, Ask for a number, Ask for a date, Ask for a URL and Buttons. Ask for a yes or a no and Ask for a rating exist too, but see above.

● Messages: Messages, Media and Goodbye message.

● Logic: Conditions, Formulas, Set a Field and Jump to.

● Integrations: Webhook, n8n, Google Sheets, Send an email and Calendly.

● AI and people: AI Agent and Human takeover.

● Code: Code and Code set.

The Calendly, Code and Code set blocks work on the web chat only.
This list grows as Landbot adds blocks, and there is no public page with the current list yet. To see it, ask your AI assistant: "Which Landbot blocks can you place?" It reads the current list from Landbot every time.
Developers can read the block catalog through the Bots API: see Get blocks.

How to describe a Landbot agent so it's built right the first time

The more specific your description, the fewer changes you need afterward. Include:

● The goal: what the agent is for, such as collecting leads or booking demos.

● The questions, in order: the exact information you need from each visitor.

● The paths: what happens after different answers, such as "if they choose Pricing, show our plans."

● The ending: what visitors see or get at the end.

● Where answers go: a Google Sheet, an email, a Calendly booking or a person on your team.

● The tone: friendly, formal or playful, and the language to use.

What summary do the Landbot MCP and the Landbot Plugin give when an agent is ready?

Every build with the Landbot MCP or the Landbot Plugin ends with a summary that includes:

● The conversation, step by step: what visitors go through, where the conversation splits and where it ends.

● Decisions you didn't make: wording, checks or settings that were chosen for you. Read this list first: it's the part you're most likely to want to change.

● Warnings: for example, a question that doesn't lead anywhere.

● Links: the link to open the agent in the Landbot builder and, once published, the share link. The Landbot MCP and the Landbot Plugin both provide these links.

What is the difference between testing and publishing?

Testing lets you try the agent privately. Publishing makes it live for every visitor.

● Test: the agent's test link shows the new version. Your visitors still see the old one.

● Publish: every visitor gets the new version of the conversation on every channel the agent uses. Design changes to a published agent are published from the Landbot builder.

● Leave it as a draft: nothing changes until you decide.

The Landbot MCP and the Landbot Plugin ask for your approval before each publish. A yes covers that one publish: after a later change, you're asked again.

How do the Landbot MCP and the Landbot Plugin test an agent?

With the Landbot Plugin in the Claude desktop app, Claude chats with your agent in a browser panel next to the conversation. It follows every path to the end with fake details, such as "Test Visitor," then checks the main path again at phone size. Each test chat appears in your Landbot inbox.

With the Landbot MCP, and with the Landbot Plugin outside the Claude desktop app, test the agent yourself on its test link or share link, following each path to the end. The Landbot Plugin lists the paths to try.

How to change an agent you already have

Ask for the change in plain words, for example: "Add a question asking for their phone number after the email." The Landbot MCP and the Landbot Plugin name the agent being changed and ask for your approval before changing it, and again before publishing it.

Don't edit the agent in the Landbot builder while the Landbot MCP or the Landbot Plugin is changing it. If a builder tab was open, refresh it before making changes, or the builder can overwrite the new changes.

How to add an AI Agent block

An AI Agent block lets your agent answer open questions with AI. Ask for it in your description, for example: "If visitors ask something else, answer from our help center, and hand over to a person if you can't."

You can teach the AI Agent what to answer from: paste text in the chat, or give it the address of a public web page, PDF or Word document. What it learns is live as soon as Landbot has read it, with no publish needed. Upload long documents in Landbot instead.
Some blocks need a paid plan. For example, the AI Agent, Conditions and Webhook blocks need the Starter plan or higher, and the Formulas block needs the Professional plan or higher. Before placing a block that needs a paid plan, your AI tool tells you which plan it needs. See Landbot pricing for what each plan includes.

How to change how the web chat looks

Describe the look you want, for example: "Make it look like our website" or "Use a dark theme with our brand blue." The Landbot Plugin writes the design with Landbot's styling guide and checks it on the share link once it's live. With the Landbot MCP, the AI assistant writes the CSS and the Landbot MCP applies it.

On an agent that was never published, the new design applies right away. On a published agent, the new design goes to the agent's draft, and visitors see it after you publish the agent from the Landbot builder. For details, see How to style your Landbot web chat with the Landbot Plugin or the Landbot MCP.

Custom CSS from the Landbot MCP or the Landbot Plugin works on paid plans and during the 14-day trial. It doesn't apply on the Sandbox plan.

💬 The Landbot MCP and the Landbot Plugin can create Web, WhatsApp, Facebook Messenger and APIchat agents. The Landbot Plugin's guided build is written for Web and WhatsApp agents. Design changes and the automatic test apply to Web only.

Was this article helpful?

What are the Landbot MCP and the Landbot Plugin?

How to fix common problems with the Landbot MCP and the Landbot Plugin

Contact