From the studio

Developer Mac memory after two days: measured

By Harish Ganapathi, Co-founder · Published November 13, 2026 · 5 min read

Most advice about Mac memory is written from a clean machine. We wanted one honest data point from a real working one. This is a single snapshot of the Mac this site was built on: not a benchmark, not an average, and not a claim about your Mac. It is a measurement with its method and its raw data attached, so you can check it and compare it with your own.

The machine and the conditions

An Apple M5 Pro with 24 GB of memory on macOS 26.5.2, up for two and a half days without a restart. Open at the time: Cursor with several project windows, a browser, a desktop AI chat app, Claude Code and Codex sessions, and Ollama installed. Docker Desktop was not running. Nothing was closed or tidied before measuring. We took the snapshot with the free Mac memory check script, which only reads system information.

What we measured

MeasureResult
Memory pressureNormal (green)
Free memory, system-wide47%
Swap used16,339 MB of 17,408 MB
Free disk space31 GB
Largest app by resident size (RSS)Cursor, about 1,772 MB across its processes
Cursor’s footprint, 54 processesAbout 11 GB, including compressed and swapped memory
Second largest appChatGPT, about 580 MB
Editor extension hosts12 processes, 1,356 MB, longest running 2.0 days
Dev servers and watchers4 processes, 825 MB, longest running 1.4 days
Processes matching MCP servers85 processes, 287 MB, longest running 2.5 days
Language servers6 processes, 159 MB, longest running 1.4 days
Ollama models loadedNone

The raw numbers are in a downloadable CSV.

What it shows

Large swap did not mean a struggling Mac. Sixteen gigabytes had been written to swap and memory pressure was still green. Swap holds memory macOS moved to disk because it was not needed. It costs you only when it is read back. Anyone who equates “swap is large” with “the Mac is in trouble” would have panicked here for no reason. The better signal is memory pressure turning yellow or red while swap grows.

RSS badly understated Cursor. The resident figure for Cursor was about 1.8 GB. The footprint macOS charges to the 54 processes whose command line points into the Cursor app, which includes memory that had been compressed or moved to swap, was about 11 GB. That is consistent with a large share of the 16 GB in swap being Cursor’s, although we did not attribute swap process by process. This is why a long-running editor can look modest in Activity Monitor’s resident column and still be the biggest thing on the machine. Our Cursor guide shows how to read the footprint figure for your own Mac.

Counting processes overstates the problem. Eighty-five processes matched an MCP server pattern, which sounds alarming. They held 287 MB in total. Most are small Node programs, and about 14 are npm wrappers around the real server, so the count is inflated. A leftover MCP server is worth stopping, but it is rarely what makes a Mac slow. The memory is in the bigger pieces.

The bigger pieces were the editor. Cursor’s resident memory was about 1.8 GB, and its extension hosts alone held 1.4 GB across twelve processes. One extension host had been running for two days. Dev servers and watchers added another 0.8 GB, with one development server holding most of it. If you want to find memory on a developer Mac, start there, not with the long tail of small helpers. Our guide on Cursor memory use covers it.

A closed app can leave a helper behind. Docker Desktop was not running, yet one Docker process had been up for two and a half days. It was a small privileged helper that holds about 2 MB, which is harmless, but it is the pattern the leftover helpers guide is about.

Method

We ran mac-memory-check.sh once. It reads sysctl kern.memorystatus_vm_pressure_level for pressure, sysctl vm.swapusage for swap, memory_pressure for free memory, and ps for per-app and per-helper memory. App figures are resident size (RSS), grouped by app: what is in RAM at that moment. RSS leaves out memory macOS has compressed or swapped out, and counts shared pages in every process that uses them, so it can read too low or too high. We also ran footprint on all of Cursor’s processes, which reports the memory macOS charges to them, including what is compressed or swapped. Helper kinds are recognised from patterns in process command lines, so a kind can include a few false matches. Only names, sizes and ages are recorded, never command lines.

What this cannot tell you

It is one Mac, measured once, under conditions we did not control. It does not compare tools, it does not show what happens on 8 GB or 16 GB machines, and it says nothing about how Cursor compares with another editor. Do not read an average into it. Memory use depends on your projects, your extensions and what else is open.

Add your Mac

A real benchmark needs many Macs. If you run the script and are willing to share the report, send it through the contact page with your Mac’s memory size, and tell us whether we may include it, in aggregate and without your name, in anything we publish. We will only publish combined results when we have enough reports to say something fair, and we will say how many there were.

If you want something that watches your Mac continuously instead of taking a snapshot, that is what Sentinely is for. The diagnosis, warnings and your first fix are free.

Questions and answers

How do I check memory usage on my own MacBook?

Open Activity Monitor and click the Memory tab, or run the free script used here. How to check memory usage on a Mac explains each number, including RAM usage and swap.

Is 16 GB of swap a problem?

Not on its own. On this Mac, memory pressure was green with 16 GB in swap. Swap costs you only when macOS has to read it back. Watch memory pressure, not the swap total.

Why did Cursor show 1.8 GB in one place and 11 GB in another?

The smaller figure is resident size: what was in RAM at that moment. The larger one is the footprint macOS charges to Cursor’s processes, which includes memory that had been compressed or moved to swap.

Is this a memory test for my Mac?

No. It is one measurement of one Mac. To test your own, run the script on your Mac and compare the report with this one.