Connecting AI Tools

Lede runs a connector that lets an AI assistant read your publication’s analytics data and answer questions about it in plain English. Instead of picking a report and setting a date range, you can ask “which articles brought in the most new subscribers last month?” and the assistant works out the query itself.

It uses the Model Context Protocol (MCP), an open standard for connecting AI assistants to outside data. Any tool that supports MCP can connect — Claude’s web app, desktop app and command-line tool all do.

What you can ask

The connector reaches the same warehouse behind the Analytics tab, but without the fixed set of reports. Questions that work well:

Which ten articles drove the most page views last month, and how long did readers spend on each?
How many subscriptions started in June, and how does that compare with May?
What share of readers who saw the paywall signed up within the same visit?
Which referring sites send readers who stay longest?
How much revenue came from customers who used a coupon, versus those who did not?
How many subscribers cancelled in the last 90 days, broken down by product?

You can also ask follow-up questions, ask it to show the query it ran, and ask it to chart or summarise the result — whatever your AI tool is capable of.

What it can and cannot do

The connector is strictly read-only. It cannot create, change or delete anything — not customers, subscriptions, products, coupons, settings or content. That restriction is enforced by the database itself, not just by convention, so no phrasing of a request can talk it into making a change.

It can read your analytics and commerce data: the event stream (page views, clicks, gate views, sign-ups, purchases, cancellations) and the supporting records for customers, products, orders, payments, sales, coupons and newsletters. It is scoped to your own publication and cannot see another publisher’s data.

Connecting

The server address is:

https://app.joinlede.com/mcp

There is no API key and no token to copy. You sign in with the same Lede account you use for the Publisher App, in your browser. If any tool asks you to paste a Lede API key, something is wrong — stop and contact support.

Claude web or desktop app

Open Settings, then Connectors.
Choose Add custom connector.
Paste https://app.joinlede.com/mcp as the URL and give it a name, such as Lede.
Click Connect. A Lede sign-in page opens in your browser.
Sign in — either with Google, or by entering your work email to receive a sign-in link. That link is single-use and expires after 15 minutes.
If you have two-factor authentication turned on, enter the code from your authenticator app.
A consent screen lists the publications the connector will be allowed to read. Check it, then click Approve.

The connector is now live. Start a new conversation and ask a question about your data.

Claude Code

From your terminal:

claude mcp add --transport http lede https://app.joinlede.com/mcp

Or add it to your MCP configuration file directly:

{
  "mcpServers": {
    "lede": {
      "type": "http",
      "url": "https://app.joinlede.com/mcp"
    }
  }
}

The same browser sign-in and approval steps follow.

Who can connect

Anyone who can sign in to the Publisher App can connect, using that same account. There is nothing extra to request and no separate credential to generate.

The connector sees exactly the publications your account covers. If you look after several titles, one connection covers all of them — you name the publication you mean when you ask a question, and the assistant can switch between them.

Lede staff accounts must have two-factor authentication turned on before they can connect an AI tool. Publisher accounts can connect either way, though we recommend turning it on. You can do that under Profile in the Publisher App.

What the assistant can see

The connector offers the assistant five capabilities, all read-only:

Read the data guide — an explanation of what data exists and how it fits together
List your publications — which titles your account covers
List tables — the data sets available for a publication
Describe a table — the columns in one data set and what type each holds
Run a query — a single read-only query against your data

You do not need to know any of this to use it. The assistant is briefed automatically on the shape of the data the moment it connects — including which figures are stored in cents rather than dollars, and how subscriptions link to payments — so plain-English questions generally work first time. If an answer looks wrong, asking it to show the query it ran is the quickest way to spot the problem.

Limits worth knowing

A query returns at most 10,000 rows and must finish within 30 seconds. The assistant is told to summarise in the query rather than pull raw rows, so this rarely bites — but a request for something like every individual page view will be cut short.
All dates and times are UTC, so a “day” runs from midnight UTC rather than your local midnight. Say so explicitly if you need a different boundary.
The data is not real-time. It is loaded into the warehouse in batches and typically runs several hours behind. Treat today’s figures as incomplete. See What is Beacon? for the detail.
Figures may not match the Analytics tab exactly. The tab caches its results for up to six hours, while the connector queries live, so the connector is sometimes slightly ahead.

Privacy and security

Your data includes customer names and email addresses. Anything the assistant reads in order to answer you is sent to whichever AI provider you have connected. Treat this as you would any other tool you put customer data into, and check it sits within your own privacy policy and your obligations to subscribers. Asking for aggregate figures rather than lists of individuals is good practice.

The connection is scoped to your publications and enforced by the database. It cannot read another publisher’s data, and it cannot write anything anywhere.
Sign-in sessions for the connector are short-lived and renew quietly in the background. Occasionally you may be asked to sign in again — that is normal.
To disconnect, remove the connector in your AI tool. If someone leaves your organisation, removing their Lede access cuts off their connector within about a minute.

If something is not working

It says your account is not allowed to see a publication. Your Lede account does not cover that title. Ask whoever administers your account, or contact support.
It finds no data at all. Your publication may not be set up for connector access yet. Contact support and we will enable it.
It asks you to sign in again mid-conversation. Expected behaviour — sign in and carry on.
A question times out. Narrow the date range, or ask for a summary rather than a full listing.

For anything else, open a support ticket from the Support tab in the Publisher App.

Was this article helpful?