AI assistants

Luftflow speaks MCP, the protocol AI assistants use to reach other tools. Connected, an assistant can build, check and dial the flows in your organization for you, through the same path the editor itself saves through.

Connect one

Every assistant adds the same server, https://luftflow.com/mcp, and signs in the first time it calls it: your browser opens luftflow, you pick the organization it may work in, and you allow it.

Claude Code

Run this once in a terminal, then sign in from /mcp in Claude Code:

claude mcp add --transport http luftflow https://luftflow.com/mcp

Codex

Add the server, then sign in:

codex mcp add luftflow --url https://luftflow.com/mcp
codex mcp login luftflow

Cursor

Use Add to Cursor on the home page, or add this to the mcpServers config Cursor reads:

{ "mcpServers": { "luftflow": { "url": "https://luftflow.com/mcp" } } }

Claude on the web or desktop

Under Customize → Connectors, add a custom connector with the address https://luftflow.com/mcp, then connect it.

Any other assistant

Add https://luftflow.com/mcp as a remote (HTTP) MCP server. It signs in with OAuth. An assistant that cannot sign in can send an API token instead: make one under Settings → API tokens, where the connector section shows the settings for your assistant with the token as a bearer credential.

What it can do

An assistant acts as you, in one organization, at your own role there. An assistant connected as an editor can change flows and publish versions, and like you it cannot put a version live unless you are an admin.

Each assistant you allow, and each token you make, is listed under Settings → API tokens. Revoke it there to cut an assistant off at once.

Working with it

Ask for outcomes: "Add a menu after the PIN screen that lists the customer's accounts from /accounts, with a Back option." The assistant reads the flow, makes the change, runs the checks, and can dial the flow to confirm the path works. Changes it makes appear in your open editor like a teammate's.