AI marketing agents for your funnels
ElasticFunnels puts your funnel in a folder of files. Claude Code or Codex reads it, builds the pages, sets up the split test, publishes it and reads the result.
Your funnel as files an agent can change
An AI marketing agent is a model that does the work instead of describing it: it reads your pages and numbers, runs commands and checks what happened.
The ef CLI binds a folder to your ElasticFunnels brand. Every page, component, script and variable becomes a file, and every action is a command with --json output and stable exit codes. That is the interface an agent needs.
What the agent can do
- Create, edit, lint and publish pages and components
- Build split tests, redirects and popups as page-event graphs
- Read revenue, conversion rate and split-test results with
ef stats - Manage products, brand variables, forms, CRM records and tags
- Put a page on a domain and list it in sitemap.xml and llms.txt
- Refuse to overwrite a change someone made in the app since its last pull
How an AI agent runs a split test
These are the commands the agent runs. We ran them on our own brand to write this page.
Pull the funnel
One command copies the brand into the folder. A session hook re-pulls before each Claude Code session.
ef pull
Build the variant
The agent writes the new advertorial or edits the old one, then lints it.
ef pages create offer/advertorial-b ef lint pages/offer/advertorial-b.ef
Wire the split test
A Split Test node in the page's events graph sends half the visitors to each variant, under the URL your ads already use.
ef pages events pull offer/advertorial ef pages events validate offer/advertorial
Publish
Both pushes refuse if the server changed since the pull, so the agent can't overwrite your edits.
ef push pages/offer/advertorial-b.ef ef pages events push offer/advertorial
Read the result
The app computes significance and flags results until each arm has its minimum sample. The agent reports that result and never computes its own; picking the winner stays with you.
ef stats splits ef stats split 526
Ask the agent how the funnel did
"Which traffic source made money this week?" becomes ef stats by utm_source. The agent reads the table and answers in plain words.
- Revenue, sessions, conversion rate, AOV and more
- Break down by page, country, device, UTM or day
--jsonfor anything the agent computes
What agents can and can't do
The agent can only run the commands the CLI has. Here's where the line is.
| Task | AI agent (CLI) | You, in the app |
|---|---|---|
| Write and publish pages | Yes: ef pages create, ef push | Or the visual builder and AI page builder |
| Same-URL split tests | Yes: page-event graphs | Or the Page Events canvas |
| Read results | Yes: ef stats, ef stats split | Or dashboards and split-test reports |
| Domains | Adds the domain and prints its DNS records | You create the records at your registrar |
| Gateways, networks, ad accounts | No | Connected in the app |
| Ad campaigns and budgets | No | You run them in the ad platform |
Claude Desktop, without a terminal
Claude Desktop can't run shell commands, so ef mcp serves your brand to it over MCP. It lists, reads, creates and edits pages, and gets preview links.
- The config file names a folder, not a secret
- Pages, components, products and preview URLs
- Split tests and stats still need the CLI
{
"mcpServers": {
"elasticfunnels": {
"command": "ef",
"args": ["mcp", "--project", "/Users/you/my-brand"]
}
}
}
Prefer clicking? Do it in the app
The built-in AI writes any page you describe, advertorials and presells included. Keep editing by chat, then split test the draft against your control.
- No terminal and no agent to install
- Runs on your AI credits, no API key
- The same pages the CLI works on
Walkthroughs with real commands
Each one shows the commands, the output and what the agent did.
Claude Code: an advertorial A/B test
Two advertorials, one URL, a 50/50 split and the result in the terminal.
AI A/B testing: the next test
Let the agent read the results and the breakdowns, then propose what to test next.
Claude Desktop MCP setup
Connect Claude Desktop to your funnels with ef mcp.
Codex CLI for marketers
AGENTS.md, the sandbox's network switch and a first page.
What the agent works on
AI marketing agent FAQ
What is an AI marketing agent?
An AI model that carries out marketing work itself instead of only suggesting it: it reads files and data, runs commands and checks the result. In ElasticFunnels that means Claude Code or Codex using the ef CLI to build pages, set up split tests, publish them and read the numbers.
Which AI agents work with ElasticFunnels?
Any agent that can run terminal commands: Claude Code, Codex and the Cursor agent. ef init writes a CLAUDE.md, an AGENTS.md and a Cursor rule that teach them the commands. Claude Desktop connects through the ef mcp server.
Can an AI agent run my ads?
No. The agent works on your funnel: pages, components, split tests, products, forms and reports. Nothing in the CLI creates, edits or pauses ad campaigns or changes budgets. ElasticFunnels sends sales back to the ad platforms, but you run the ads.
Can Claude Desktop set up a split test?
Not through MCP today. The ef mcp server gives Claude Desktop eight tools: list, read, create and edit pages, read components and products, and get preview links. Split tests and stats need the CLI, so use Claude Code or Codex for those.
Do I paste an OpenAI or Anthropic API key into ElasticFunnels?
No. The agent runs on your own Claude or ChatGPT account. ef init signs in to your brand through the browser and keeps the credential in the project folder, so the agent never sees a key in chat.
What still needs the ElasticFunnels app?
Connecting payment gateways, affiliate networks and ad accounts, and creating the DNS records for a new domain at your registrar. The agent can add a domain and print its DNS records, but you set them.
Is it included on every plan?
Yes. The ef CLI, the MCP server and the in-app AI page builder are included on every plan. Plans differ only in usage limits.