SaaSToolkit MCP is live: ask Claude, ChatGPT or Cursor about your customers

Picture a normal Tuesday check-in... You open analytics, then the email tool, then the support inbox, then the ad dashboard, and somewhere around the fourth tab you forget what you were trying to find out?!
Yesterday we launched the SaaSToolkit MCP, which lets Claude, ChatGPT, Cursor or any other MCP client read your SaaSToolkit projects directly. You ask a plain question and the answer comes from your tracked data. It’s included on every plan, the free Starter plan too!
This post covers what MCP is, what our server can and can’t do, how to connect it, how the Product Hunt launch went (honestly), and why we think it’s worth setting up.
MCP in plain English
MCP is short for Model Context Protocol. Anthropic introduced it in November 2024 as an open standard for connecting AI applications to outside tools and data. The easiest way to picture it is one agreed plug shape: a product builds a single MCP server, and any assistant that speaks MCP can connect to it, with no custom integration for each one.
It caught on VERY quickly. In December 2025 Anthropic donated MCP to the Agentic AI Foundation, a fund under the Linux Foundation that Anthropic, Block and OpenAI co-founded, with Google, Microsoft, AWS, Cloudflare and Bloomberg among its supporters. Anthropic was reporting more than 10,000 active public servers by then, and by spring 2026 the SDKs were being downloaded roughly 97 million times a month. (Download counts are a rough proxy for real use, so read that as direction rather than a headcount.)
We think more business questions are going to start in an AI assistant than in a dashboard. If your product can’t be reached from there, you’re asking people to leave the place they already work. Our founder Ian Naylor put it this way on the launch page: “If your AI can’t see your customers, it’s working blind.”
If you want the background on asking your data questions in plain English or on agentic analytics, we’ve written about both. The MCP is where those ideas meet the AI tool you already have open.
What the SaaSToolkit MCP does
Your single tracking snippet already links anonymous visitors to paying customers under one identity. The MCP gives your AI assistant a way to read that whole picture: traffic, funnels, revenue, ad spend, contacts, email, support and saved AI insights.
There are 39 tools. 30 of them only read data. The other 9 can write, and we’ll cover exactly what those can and can’t do in a minute.
Area | What your assistant can do | Example tools |
|---|---|---|
Projects and traffic | List your projects, pull headline visitor numbers, daily trends, and breakdowns by country, device, channel, referrer, UTM or entry page |
|
Pages, funnels and events | Show landing page performance, step-by-step funnel drop-off, and signups and purchases per page by device |
|
Revenue and ads | Report daily revenue, ad spend, clicks and conversions by platform and campaign, and how much of your traffic carries UTMs and click IDs |
|
Leads and customers | Find a contact, show their full timeline of visits, events, emails, support and purchases, and list contacts in a churn-risk score range |
|
Report campaign sends, opens, clicks, replies and bounces, read inbound replies, reply in thread, and draft an automation |
| |
Support | See chats and tickets in one queue, read full threads, reply as a human agent, and change status or priority |
|
Phone (Website SDR add-on) | List phone campaigns and calls with outcome, duration, sentiment and summary |
|
Saved AI insights | Read trends, warnings and recommendations for one project or across all of them |
|
What it looks like in use
These are the questions we put on the setup page, and they make a good first hour:
Which channels, campaigns and landing pages bring visitors who sign up and pay?
Look up a customer by name or email and show their full timeline.
Which contacts have a high churn-risk score, and what did they do last?
Which support chats need a reply? Show me the conversation.
How did each email campaign perform, including replies?
Show the saved insights for my projects.
For a first test, ask it to list your projects, then report your main project’s visitors over the last 7 days and where they came from. If that comes back right, the connection works.
Take the first question. Answering it properly means pulling traffic by channel and landing page, then lining that up with signups and purchases, which is work for tools like get_traffic_breakdown and get_page_device_conversion. You don’t pick the tools. Your assistant does, and it shows its working in the chat.
The conversation doesn’t have to stop at the answer, either. Ask which support chats are waiting, read one, ask for a draft reply and tell your assistant to send it. The reply goes out as a human agent, the SaaSToolkit AI assistant pauses on that chat so the two don’t talk over each other, and if the visitor has left the site they get the reply by email.

What it can’t do, on purpose
An assistant that can send things needs hard limits, so here they are.
The 9 write tools can reply to support chats, tickets and email replies, change a chat’s or ticket’s status or priority, mark an email reply as read or move it to Junk (nothing is deleted), create a draft email automation, and send test emails capped at 5 per project per hour.
They can’t switch an automation on, send a campaign to your contacts, or change your settings, billing or contacts. Any automation your assistant builds is saved switched off and never enrols existing contacts, so a mistaken email to your whole list can’t happen from here.
Sign-in uses OAuth with your SaaSToolkit account, so there’s no API key to copy, paste or leak, and the server only sees projects you can already access under your team permissions.
Your AI tool adds its own controls on top. Claude, ChatGPT and Cursor let you switch individual MCP tools on or off, and many ask before running a write action. That means you can keep every read tool and the drafting tools on, and switch off the one that changes a chat’s status if you’d rather do that yourself. To disconnect, remove SaaSToolkit from your tool’s connectors.
Connecting it
The server URL is https://mcp.saastoolkit.ai/mcp (Streamable HTTP, OAuth sign-in). You need a SaaSToolkit account with at least one project. The AI tool registers itself, opens a browser window, you sign in and approve access.
Claude on the web or desktop: go to Customize, then Connectors, then Add custom connector. Name it, paste the URL, choose Register automatically if asked, then Connect. Custom connectors depend on your Claude plan.
ChatGPT (web): in Plugins, select + and then Add custom MCP server. Name it, paste the URL, choose OAuth, create it as a plugin and sign in. Then type @ in a chat to pick it. In a work workspace, an admin may need to allow custom MCP servers first.
If you work in a terminal, these are the commands:
# Claude Code (then run /mcp, choose saastoolkit and select Authenticate)
claude mcp add --transport http --scope user saastoolkit https://mcp.saastoolkit.ai/mcp
# Codex
codex mcp add saastoolkit --url https://mcp.saastoolkit.ai/mcp
codex mcp login saastoolkit
# Gemini CLI (then run /mcp auth saastoolkit)
gemini mcp add --transport http --scope user saastoolkit https://mcp.saastoolkit.ai/mcp
# opencode
opencode mcp add saastoolkit --global --url https://mcp.saastoolkit.ai/mcp
opencode mcp auth saastoolkitCursor, VS Code with Copilot Chat in agent mode, and Lovable are covered too (Lovable works in project chat only, not in published apps). Any other tool that supports remote MCP over Streamable HTTP with OAuth and automatic client registration should connect the same way. The full steps are on the MCP setup page.
The two phone tools need the Website SDR add-on. Everything else works on every plan.
The Product Hunt launch
We launched on Product Hunt on 18 November, listed as SaaS Toolkit MCP, with a tagline pitching it as an SDR and CMO on your site around the clock for one JavaScript snippet. It was a busy day to launch an MCP: Product Hunt now lists MCP as a topic and a good share of the day’s products carried the tag.
Ian’s launch comment started with his first SaaS, 14 years ago. The hard part, he wrote, was never building the product. It was knowing which visitors were worth talking to, who was about to leave and which support chat needed him right now, when the answers were spread across too many dashboards to open. The MCP is that problem handed to your AI agents. He closed with two questions for anyone passing through: what’s the first thing you’d ask your AI about your business, and what should it be able to do next?
As I write this on 19 November, the launch has 73 upvotes, sits at number 27 for the day, and the product page has 45 followers. That’s a mid-table finish, which isn't bad for a new product and we can hold our heads high and say we did not ask even a single person to go and vote for us to up the number count!
The part worth reporting is the conversation. The sharpest question came from a commenter who asked whether the write tools could be switched off individually, so an assistant could keep drafting replies but never change a chat’s status on its own. They can: Claude, ChatGPT and Cursor all let you toggle individual MCP tools, and nothing is sent or changed unless you ask your assistant to. That answer now lives in the section above because someone asked for it.
Why it’s worth connecting
One question can cross every part of the business. Traffic, revenue, ad spend, contacts, email and support sit behind one login, so “which contacts have a high churn-risk score, and what did they do last?” gets an answer that touches contacts, timelines and support in a single reply. That’s the work that used to need four tabs.
You stay in the tool you already use. Claude, ChatGPT, Cursor, Codex, Gemini CLI, VS Code, Lovable and opencode all work. There’s no new dashboard to learn and no login to remember.
Several projects, one question. get_portfolio_insights reads saved insights across every project you have access to. If you run more than one product, or look after sites for clients, that’s one prompt instead of one login per project.
You can act on the answer. Reading a support thread, drafting a reply and sending it happens in the same conversation, and the tools that change things are the nine you can see listed above.
The limits are visible and yours to adjust. Read-only by default (30 of 39 tools), automations saved switched off, no campaign sends, OAuth with no key to leak, and per-tool switches in your AI client.
It’s also included. Every plan has it, including the free Starter plan.
Try it
Connect it, then ask the question you’ve been putting off. We’d love to hear what you asked and what you wish it could do next, and the Product Hunt page is the easiest place to tell us. Setup details live on the MCP setup page.