MCP server

Connect Claude, ChatGPT, Codex and other AI assistants to your Landbot account, so they can build, edit, test and publish your bots for you.

The Landbot MCP server lets an AI assistant work with your bots through the Model Context Protocol. You describe the bot you want in plain words, and the assistant reads the block catalog, creates the bot, wires its blocks, and deploys it to test or publishes it once you say so.

https://mcp.landbot.io/mcp

It uses the Streamable HTTP transport and OAuth: there is no token to copy. The first time you connect, your client opens Landbot's login (password or Google) and a consent screen. From then on the assistant acts with your account's permissions, on your brand. If you don't have a Landbot account yet, you can sign up from that same screen.

Note This is not the docs MCP at https://dev.landbot.io/mcp described in For AI agents, which only searches and reads these docs.

Supported clients

Client How to connect
claude.ai Custom connector
ChatGPT Connector in developer mode
Claude Code claude mcp add
Codex codex mcp add

claude.ai

  1. Go to Settings → Connectors → Add custom connector.
  2. Name it Landbot and paste https://mcp.landbot.io/mcp.
  3. Click Connect, log in with your Landbot account and approve.

On Team and Enterprise plans, an Owner adds the connector once for the whole organization. Each member then finds Landbot under Settings → Connectors and clicks Connect.

ChatGPT

  1. Go to Settings → Apps & Connectors → Advanced settings and turn on Developer mode.
  2. Back in Apps & Connectors, click Create, name it Landbot, paste https://mcp.landbot.io/mcp as the MCP server URL and pick OAuth for authentication.
  3. Click Create, log in with your Landbot account and approve.

Claude Code

claude mcp add --transport http landbot https://mcp.landbot.io/mcp

Then run /mcp in Claude Code, pick landbot and log in.

Codex

codex mcp add landbot --url https://mcp.landbot.io/mcp
codex mcp login landbot

Permissions

The consent screen asks for these scopes:

Scope Allows
bots:read List and read your bots, their drafts and the block catalog
bots:write Create and edit bots, upload the files they show, deploy them to test and publish them
agents:read Read the AI agents your bots run, what they know and the APIs they call
agents:write Create, edit and publish the AI agents your bots run, teach them texts, web pages and documents, and set the APIs they call
integrations:read List the accounts connected to your integrations, the spreadsheets, tabs and columns a Google account reaches, and a Calendly account's event types
integrations:write Ask for another account to be connected, through a link you open in your browser
channels:read Read how your web chats look
channels:write Change how your web chats look and the texts they show
scripts:write Run your own scripts on your web chat pages

What it can do

Group Tools
Read list_bots, get_bot, get_bot_draft, list_blocks, get_block_definition, get_guide (how to build a bot step by step, how to style its web chat, and how to hand it back and try it)
Build create_bot, update_bot, add_draft_blocks, update_draft_block, delete_draft_block, apply_draft_changes, replace_draft
AI agents get_ai_agent_schema, get_ai_agent, create_ai_agent, save_ai_agent_draft, list_knowledge_sources, add_knowledge_text, add_knowledge_website, create_knowledge_document_upload, list_ai_agent_actions, save_ai_agent_action, delete_ai_agent_action, get_ai_agent_credentials
Integrations list_connected_accounts, start_account_connection, get_account_connection, list_spreadsheets, list_sheets, list_sheet_columns, list_calendly_event_types
Web chat look get_web_chat, update_web_chat (Custom CSS, Custom JS as custom_js, the v4 web chat, the Back button, the header title and avatar, the tab title, the hints of the text box and the buttons)
Media create_media_upload
Test and publish deploy_draft_to_test, publish_bot, publish_ai_agent

Tools that change something ask for your approval in the client, and the assistant only deploys a bot to test or publishes it after you say yes.

Google Sheets, SendGrid and Calendly blocks go through an account connected to your brand. When none will do, the assistant gives you a link: open it in your browser, signed in to Landbot, and consent at Google or Calendly, or enter the SendGrid sender and API key there. Your passwords and keys never pass through the conversation.

An AI agent calls your own APIs through its REST API actions. The assistant writes each action and names the headers that carry a credential, such as an API key; you enter their values in Landbot, from a link the assistant gives you, so they never pass through the conversation either. Actions apply at once, without publishing the agent, to the conversations that reach it afterwards. Moving an action to another scheme, host or port empties its credentials, and you enter them again.

An AI agent learns from texts, from public web pages, including a link to a PDF or a Word document, and from documents on your computer. A document on your computer reaches an AI agent through create_knowledge_document_upload: the assistant gets a link that works for 15 minutes and sends the document to it straight from your machine, so it takes a client that can run commands there, such as Claude Code or Codex. It reads PDF, Word (.docx), plain text, Markdown, RTF and HTML documents of up to 5 MB. In a client that cannot, a file you attach in the conversation goes in as text, so it suits short documents; upload a longer one in Landbot.

An image, a video, an audio or a document on your computer reaches a bot through create_media_upload: the assistant gets a link that works for 15 minutes and sends the file to it straight from your machine, so it takes a client that can run commands there, such as Claude Code or Codex. Files go up to 30 MB, and once uploaded anyone with their link can open them. A file that is already online needs no upload: the assistant puts its link in the block.

Example prompts

List my Landbot bots.
Build a Landbot bot that greets the visitor, asks for their email and says goodbye.
Add a rating question at the end of my bot "Support", then deploy it to test.
Build a bot that asks for a name and an email and saves them as a new row in my Google Sheet.
Teach the AI agent of my bot "Support" what https://example.com says.