Create a Chatbot with ChatMaxima MCP

Build a full ChatMaxima chatbot from Claude, Gemini or ChatGPT. The assistant creates the bot, adds and connects builder blocks, organizes the canvas and runs a test.

Overview

With ChatMaxima MCP connected, your AI assistant can build a complete chatbot in your workspace from a plain-language brief. The assistant uses the real ChatMaxima builder, so every block it creates is a genuine builder block that you can open, edit and publish like any bot you built by hand. There is no template extraction and no drift from the live builder.

This page explains how chatbot building works through MCP and how to get the best results.

Prerequisites

Before you begin, ensure you have:

A connected AI client. See Connect with Claude, Gemini or Codex / ChatGPT.
Builder access in your role. Your ChatMaxima role must allow the chatbot builder.
An active plan. A Free or expired workspace cannot save builder changes.

How the assistant builds a bot

The assistant follows the same flow a person would in the visual builder:

Discover the blocks. It calls list_block_types to see the block catalog and get_block_schema to learn the exact fields each block needs.
Create the bot. It calls create_bot to make a new, empty bot in your workspace.
Add blocks one by one. It calls add_block for each step. Each new block is threaded after the previous one, so the conversation flows in order. The canvas is auto-organized as blocks are added.
Branch where needed. For buttons or conditions, it calls connect_blocks to wire a specific output to its target block.
Tidy and verify. It calls organize_bot for a final clean layout, then run_test with a scripted conversation to confirm the bot behaves as intended.

You can watch the result appear in the dashboard builder once the assistant finishes.

Step 1: Describe the bot you want

Give the assistant a clear brief. The more specific you are, the closer the first build will be. A good brief includes:

The goal of the bot (qualify leads, book appointments, answer FAQs).
The industry or audience so the wording fits.
The questions or steps the bot should cover.
The outcome at the end (create a lead, hand off to an agent, send a link).

Example prompt:

Using ChatMaxima, create a lead qualification chatbot for a real estate agency.
Greet the visitor, ask whether they want to buy or rent, capture their name,
phone number and preferred area, then thank them and hand off to a human agent.

Step 2: Let the assistant build and connect blocks

The assistant will create the bot and add blocks such as a greeting message, question blocks, button choices and a handoff. For any branch (for example a "Buy" button versus a "Rent" button), it wires each button output to the right next block with connect_blocks.

Common blocks the assistant uses

Block	Purpose
Message	Send text, media or a greeting
Question	Capture a typed answer into a variable
Buttons	Offer choices that branch the conversation
Condition	Route based on a value or attribute
Handoff	Pass the chat to a human agent

Note: Always ask the assistant to call get_block_schema before adding a block type it has not used yet, so it supplies the correct fields. Most clients do this automatically.

Step 3: Review the test run

The assistant calls run_test with a sample conversation and reports how the bot responded. Read this back to confirm the flow makes sense. If a branch is wrong or a question is missing, ask the assistant to fix it. It can update a block with update_block, remove one with delete_block, add new blocks, or re-run the test.

Step 4: Open and publish in ChatMaxima

Once you are happy, open the bot in the ChatMaxima dashboard builder. Everything the assistant created is editable there. Make any final touches, then publish or attach the bot to a channel as you normally would.

Best Practices

Build in stages. Ask for a first version, review the test, then request changes. This is faster than one huge brief.
Name your variables. Tell the assistant what to capture (name, phone, email) so the data lands in the right place.
Keep branches simple. Two or three button choices per step read better than many.
Verify before publishing. Use the test run as your acceptance check.

Troubleshooting

The assistant says it cannot save the bot

Confirm your workspace plan is active. A Free or expired plan blocks builder saves.
Confirm your role has builder access.
Reconnect the client and make sure you selected the correct workspace on the consent screen.

The bot tools are not listed

Your role may not include builder access. Ask a workspace admin to grant it.
Reconnect the client so it refreshes the available tools.

A branch goes to the wrong place

Ask the assistant to re-check the connections and re-wire the button or condition output.
Ask it to run the test again to confirm.