Clarflow MCP Server

Connect Clarflow to Claude, Cursor, or any AI tool that speaks the Model Context Protocol and manage your whole workspace in plain language — build a quiz funnel, publish it, read the drop-off, launch an A/B test.

There is nothing to install. Clarflow hosts the server; you add a URL and an API key.

• Server URL — https://mcp.clarflow.com
• Transport — Streamable HTTP
• Authentication — Authorization: Bearer <your API key>

Use the www. host. The apex clarflow.com redirects, and most HTTP clients drop the Authorization header when a redirect crosses hosts.

What you can do

Once connected, you can ask for things like:

• "Build me a 6-step skincare quiz that segments by skin type, then publish it."
• "Which step of my supplement quiz is losing the most people?"
• "What are people actually answering on question 3?"
• "Set up a 50/50 A/B test between my two landing quizzes."
• "Add three more questions to the end of my collagen funnel."

The quick way

1. Open Clarflow and go to Settings → API & MCP.
2. Pick your tool — Claude Code, Cursor, Claude Desktop, or something else.
3. Click Connect.

Clarflow hands you the finished setup with your key already in it: a single command to paste for Claude Code, a one-click Add to Cursor button for Cursor, and the exact two values to paste for Claude Desktop. Nothing to splice together.

The setup is shown once, because Clarflow stores only a hash of your key and cannot show it again. If you lose it, hit Reconnect on the connection and you'll get a fresh one.

To disconnect, click Remove connection — it stops working immediately.

Everything below is the manual version, if you'd rather wire it up yourself or you're using a tool that isn't listed.

What a connection can do

A connection is a personal credential: it acts as you, in one workspace, with whatever permissions you have right now.

Because the role is read live on every request, a connection automatically loses access when you are removed from the workspace or your role is reduced. There is nothing to clean up.

A connection can never: touch billing, manage members, or reach any other workspace.

Connecting manually

Clarflow fills your key into all of these for you in Settings → API & MCP — these are here for reference, or for tools not in the list. Replace cf_live_your_key_here with your own key.

Claude Code

claude mcp add --transport http clarflow https://mcp.clarflow.com \
  --header "Authorization: Bearer cf_live_your_key_here"

Then confirm it registered:

claude mcp list

Add --scope user to make it available in every project, or --scope project to share it with your team through the repo's .mcp.json (do not commit a literal key — see the note below).

Cursor

The Add to Cursor button in Clarflow installs this for you in one click. To do it by hand, add to ~/.cursor/mcp.json for all projects, or .cursor/mcp.json inside one project:

{
  "mcpServers": {
    "clarflow": {
      "url": "https://mcp.clarflow.com",
      "headers": {
        "Authorization": "Bearer ${env:CLARFLOW_API_KEY}"
      }
    }
  }
}

Then set CLARFLOW_API_KEY in your environment. Cursor resolves ${env:…} in mcp.json, which keeps the key out of a file that can easily end up in version control. (Clarflow's one-click button writes the key inline, since it can't set an environment variable for you — use this form instead if the file is shared.)

Claude Desktop

Open Settings → Connectors → Add custom connector, then enter:

• URL — https://mcp.clarflow.com
• Header — Authorization: Bearer cf_live_your_key_here

Any other MCP client

Point it at the URL with the bearer header. Clients that cannot speak Streamable HTTP directly can bridge through mcp-remote:

{
  "mcpServers": {
    "clarflow": {
      "command": "npx",
      "args": [
        "-y", "mcp-remote",
        "https://mcp.clarflow.com",
        "--header", "Authorization: Bearer cf_live_your_key_here"
      ]
    }
  }
}

Tools

The A/B testing tools only appear if cross-funnel A/B testing is enabled on your account. If you don't see them, contact support.

Resources

Two references are also exposed as MCP resources, for clients that support them:

• clarflow://schema/funnel-json — the funnel JSON format
• clarflow://schema/css-classes — every CSS class a funnel exposes

A worked example

You: Build me a 6-step quiz that helps people find the right protein powder, then publish it.

The assistant will typically:

1. Call get_workspace_info to see what it's allowed to do.
2. Call get_funnel_schema to read the funnel JSON format.
3. Call create_funnel with the whole quiz.
4. Call get_funnel to check the result reads well.
5. Call get_funnel_changes — it will report first_publish.
6. Call publish_funnel and hand back the live URL.

You: A week later — where is that funnel losing people?

7. get_funnel_analytics returns per-step drop-off and names the worst step.
8. get_question_responses shows what people picked before they left.

Working with funnel content

Funnels are built from a JSON format called clarflow-canvas-nodes — the same format the Clarflow builder accepts when you paste onto the canvas. The full reference lives at /docs/json-builder, and the assistant can fetch it itself with get_funnel_schema.

One thing worth knowing, because it trips people up: get_funnel with include: "full" returns the stored format (nodes keyed by id, edges using from/to), while the write tools take the clipboard format (arrays, edges using source/target). They are not interchangeable. If you are editing an existing funnel, prefer update_funnel with mode: "append" — it generates ids for you and never round-trips the whole document.

Safety

The assistant is driving your live account, so a few things are worth knowing before you start.

Publishing is immediate. publish_funnel puts a funnel in front of real traffic.

Structural changes reset analytics. If you change a funnel's steps, its existing results can't carry over — publishing archives them to a version and restarts collection from zero. The API refuses this until it is asked for explicitly, so the assistant has to confirm with you first.

Deleting is permanent. delete_funnel removes the funnel and its analytics with no undo.

Publishing affects live experiments. If a funnel is a variant in a running A/B test, publishing changes that experiment mid-flight. The response says so.

Your builder tab wins by default. If someone has the funnel open in Clarflow, a write through the API is refused rather than being silently overwritten by their next autosave.

Destructive tools are marked as such in the protocol, so clients like Claude Code will ask before running them. You can Remove connection at any time from Settings → API & MCP; it stops working on the next request.

Troubleshooting

Support

Questions about the MCP server: support@clarflow.com.