
fnlowl
com.fnlowlv1.0.1Updated Oct 2, 2026
Read and manage fnlowl widgets, leads, email sequences and analytics from an AI assistant.
Overview
Lets an assistant read and manage fnlowl lead-capture widgets, leads, email sequences and funnel analytics through a hosted MCP endpoint.
- What it does
- Connects an assistant to fnlowl's hosted MCP surface for lead-capture widgets, the leads they produce, and the follow-up mail. Documented skills cover building and diagnosing widgets, funnel analytics, drafting and reviewing campaigns and sequences, and installing embeds on generic sites, WordPress, Shopify, Webflow and Framer. A list_projects tool returns project ids for scoping calls.
- When to use it
- Useful when you already run fnlowl funnels and want an assistant to inspect widget performance, work with leads, or draft follow-up sequences and embed installs without switching to the fnlowl UI.
- Requirements
- Remote streamable HTTP endpoint at OAuth sign-in and scope approval on first use; tokens are revocable under Settings, API, Connected apps. Scripts and CLI can instead use an API key sent as Authorization: Bearer. Optional x-fnlowl-project header scopes calls to one project.
Installation
In SourceWeft
- Open fnlowl in the dashboard and add it to a workspace.
- Enable the server for the chats that should use its tools.
Web executable via Streamable HTTP. Remote servers run from the web runtime once configured in a workspace.
Other MCP clients
Add this to your client's mcpServers config.
{
"mcpServers": {
"fnlowl": {
"type": "http",
"url": "https://api.fnlowl.com/v1/mcp"
}
}
}README
@fnlowl/skills
Agent skills for operating fnlowl (pronounced Funnel Owl — Funnel Intelligence, Agentified!) — lead-capture widgets, the leads they produce, and the mail that follows.
Four skills:
fnlowl-capture— build and diagnose widgets, leads, and funnel analytics.fnlowl-follow-up-automation— draft and review delivery, integration handoff, campaigns, and sequences.fnlowl-installation— install embeds on generic sites, WordPress, Shopify, Webflow, and Framer.fnlowl-mcp— connect an agent to the remote MCP surface safely.
Read them in skills/.
Install
Name your agent with --agent (for example claude-code, codex or cursor). Without it, a
non-interactive run (--yes) installs into every agent the installer supports.
Pin a release
Releases are tagged on GitHub. For a stable install, pin a tag instead of the default branch:
The skills are versioned together as one bundle. A release that renames or removes a skill says so in its notes.
Install as a Claude Code plugin
The repo is also a plugin marketplace, so Claude Code can install it directly:
How to invoke skills
Verify install
Try a task that should trigger one of the skills:
- "Create a quiz widget for my pricing page."
- "Why is this widget not converting?"
- "Draft a three-email nurture sequence for people who finish the quiz."
- "Add the fnlowl widget to my WordPress site."
- "Connect Claude to fnlowl and list the tools it can use."
What's included
- four skills that load when relevant
- reference files and recipes for the capture and follow-up skills
.mcp.json, which points an agent at the hosted MCP server (OAuth sign-in on first use)
It does not ship slash commands or an MCP server of its own; the server runs at https://api.fnlowl.com/v1/mcp.
Use it
The MCP server is at https://api.fnlowl.com/v1/mcp (streamable HTTP, OAuth). .mcp.json in
this package points there. Your client opens fnlowl's consent screen the first time: sign in and
approve the scopes and website. Tokens are fnla_, revocable under Settings → API →
Connected apps; a connector limited to one website is already project-bound, so x-fnlowl-project is
ignored.
Claude.ai and ChatGPT work the same way: add the URL as a custom connector (OAuth, leave client id and secret blank).
For scripts and the CLI you can use an API key from Settings → API instead. It starts fnlk_,
is shown once, and is sent as Authorization: Bearer fnlk_....
Scope calls to one project with the x-fnlowl-project header. list_projects
returns the ids.
What it will not do
The remote MCP surface has no campaign or transactional send tool and cannot unsubscribe anyone. Publishing widgets and activating sequences are separate confirmation-required actions; the server cannot verify that conversational confirmation. High-risk Composio connection/action tools are withheld when that optional integration is configured. See references/approval-model.md.
Develop
Markdown and JSON, no dependencies — Node 18 or later. The one rule is describe only what is built, and a limitation goes in references/known-gaps.md rather than being left out.
The package also includes Codex, Claude, Cursor, and agent-marketplace plugin manifests.
License
MIT — see LICENSE.
Repository layout
Source: README.md at commit c4eaf20
Tools
0Version history
1- v1.0.1LatestOct 2, 2026
