Using Cursor and MCP as a Product Manager

Cursor is an AI-powered code editor built for software engineers – not product managers. Yet over the past few months, it has become my primary AI tool for product management work. Here’s why I made the switch, and how it’s helping me get more done.

A Case Study (and Some Use Cases)

Here’s a recent example of how I used Cursor end-to-end to investigate and document an analytics issue:

Exploring our analytics. I can chat directly with PostHog (our analytics tool) through Cursor. It writes and executes SQL queries, then presents the results to me in plain English.
Identifying the root cause of issues. I noticed something unusual in our analytics data. When I asked Cursor to investigate, it autonomously created multiple diagnostic queries, pinpointed the exact date the bug began, and identified the most likely causes.
Drafting markdown documentation. Cursor summarised the issue in a markdown file. Like most LLM outputs, it was verbose. I ended up deleting about half the content. But it gave me a solid starting point.
Publishing to Notion. Once I refined the writeup, Cursor posted it to Notion and automatically applied the relevant tags – something most people don’t do consistently! I manually added an “AI generated content” tag-transparency about AI involvement matters to me.
Creating a Linear ticket. Cursor created the ticket with a concise description, and assigned it to the correct team and owner.

This end-to-end workflow took minutes to go from investigation to documentation to ticket creation. If I was doing this all myself it would have taken hours. More importantly, it all happened without me leaving Cursor.

Another case study (and some more use cases)

Here’s how I used Cursor to analyse historical data on a specific user segment and produce a comprehensive report.

Searching and summarizing Notion documentation. I needed to compile feedback from a specific user segment. Cursor searched both our user feedback and meeting notes databases in Notion. Surprisingly, it surfaced documents I didn’t know existed, though it also missed some information I was aware of. Not perfect, but a strong 80/20 solution.
Analysing feedback. I had hundreds of survey responses from this segment. After removing PII from the comments, I asked Cursor to design a classification scheme, apply it to all responses, then produce both quantitative and qualitative analysis. It handled the entire analytical workflow autonomously. Some of the classifications I didn’t think looked quite right, but it was easier to edit something that was mostly correct than it would have been to create it all myself from scratch.
Generating data visualisations. Within seconds, Cursor created a beautiful HTML bar chart visualising the key findings. I simply took a screenshot and dropped it into my final report.
Consolidating all analysis into a final report. Cursor identified patterns across all the research data. This is where LLMs excel. While my final report looks nothing like Cursor’s first draft (I rejected most suggestions), it surfaced valuable statistics and observations I incorporated into my analysis.

This one still took me a couple of hours to complete. But without Cursor, it either would have taken a lot longer or have been done to a lower standard.

Other use cases

Writing status updates. For project updates (posted in Linear and synced to Slack), I start with Cursor’s summary of progress since my last update. I heavily edit these drafts-cutting most of the content. But it prevents me from forgetting important details and eliminates the need to manually review dozens of tickets. I let Cursor be comprehensive, then I ruthlessly prioritise what stakeholders actually need to know.
Proofreading. Cursor’s track changes feature lets me review and approve/reject each suggested edit individually. Combined with a spell-check extension, it’s become my default proofreading workflow.
Critiquing my work. I’m building a library of prompts that simulate feedback from different expert perspectives (inspired by the Cursor for Product Managers course). This helps me strengthen strategy documents before sharing them with the team.

The Power of MCP

Connection with third-party tools. MCP (Model Context Protocol) enables Cursor to connect directly with Notion, Linear, and PostHog. This isn’t just about eliminating copy-paste. Cursor can query these data sources directly, access the full context it needs, and write back to create documentation, tickets, and analytics insights. It’s the difference between an assistant who reads screenshots versus one with direct database access. It also means that rather than switching between all these tools, I can stay working in one place: Cursor.

Here are the MCP servers I am using at the moment. You can control which actions to give Cursor permission to do, and whether or not to approve each individual request to use a tool in a particular way

How I stay in control

Using AI extensively requires deliberate guardrails. Here’s how I maintain control:

AI generates, I decide. Every edit goes through track changes. I review and approve each modification individually. Nothing changes without my explicit approval.
Heavy editing is the norm. I typically cut more than 50% of AI-generated content. Cursor gives me comprehensive drafts; I refine them to what’s actually important.
Transparency about AI use. I label AI-generated content in shared documents. My team should know when they’re reading AI output versus my direct work.
Human judgment on strategy. I use AI for research, analysis, and drafting. But strategic decisions remain mine. AI helps me think better; it doesn’t think for me.
Local storage and version control. All work lives in markdown files on my machine. I can see the full history of changes and revert anything if needed.

What Cursor Does Better Than ChatGPT (& Other Similar Tools)

Outputs are produced in markdown and stored locally. Everything lives in markdown files on my machine – not trapped in a chat interface. This means I can directly edit AI outputs, version control them, and iterate without copy/pasting. Markdown is simple, portable, and supports the basic formatting I need.
Track changes functionality. Cursor uses the same diff interface engineers use for code review. Every AI suggestion appears as a tracked change I can individually accept or reject. This gives me confidence that nothing changes without my explicit approval. I don’t have this same confidence when using ChatGPT.
Multiple AI models. Cursor supports multiple AI models in one interface. As model capabilities evolve and AI leaderboards shift constantly, I can switch to whichever model performs best for my current task. No need to maintain separate subscriptions or learn different interfaces.
One subscription for everything. A single Cursor Pro subscription has the potential to cover all my AI needs at work – product management, light coding, analysis, searching the web and writing.

*An example of how Cursor helped me proofread this blog post – and approve each individual change it makes to my document*

Other things Cursor does

Cursor also offers features common to tools like ChatGPT, but worth mentioning:

Always considers my product strategy and writing style automatically in every response (via indexed project files).
Easy prompt template storage via @ commands. No more maintaining a separate prompt library and copy-pasting-templates live in my project and are accessible with autocomplete.
All chats stored for easy reference and continuity. I can pick up exactly where I left off in any previous conversation.
Zero data retention policy. AI models don’t train on my company data-whether from chat or accessed via MCP. Note: This guarantee only applies when using Cursor’s included LLM credits, not when bringing your own API keys.
Intuitive interface. Folder structure left, files centre, chat right. The layout feels natural. I considered alternatives like Claude Code (terminal-based), but Cursor’s visual interface felt easier than using terminal.

Weaknesses of Using Cursor as a product manager

MCP servers frequently disconnect and require re-authentication. This can be frustrating when you’re in the middle of a complex task.
No cross-session context memory. Unlike ChatGPT’s memory feature, Cursor doesn’t automatically learn from past conversations to understand my preferences and context. Chat history is saved locally, but the AI doesn’t build a persistent understanding of me across sessions. This may be coming in a future update, but for now, each conversation starts relatively fresh.

How you can get started

If you want to get started with using Cursor yourself or you want to learn more, here are the two resources I recommend:

Cursor for Product Managers: Learn Cursor IN Cursor!This brilliant free introductory course teaches you how to use Cursor, in Cursor. As well as being an excellent course, this is a fascinating insight into the potential for AI tooling to support good course design.
This podcast from How I AI is what first got me interested in the idea of using Cursor as a product manager: “Cursor is a much better product manager than I ever was”: How this PM uses AI for PRDs, Jira tickets, and replying to coworkers | Dennis Yang (Chime).

Recommended extensions for Cursor

Spell Checker – Catches typos in real-time as you write markdown documents.
Editable Markdown Preview – Allows you to edit markdown whilst seeing how it will look.