Excel MCP Server usage statistics¶
Excel MCP Server lets GitHub Copilot, Claude, and other AI assistants automate the real Microsoft Excel application. This public report shows how the open-source project is used and where people get it.
New to the project? Install Excel MCP Server to get started.
Anonymous public report
This page shows broad usage patterns, not individual activity. Names, file details, locations, and the content of workbooks are never included.
Last updated: Oct 12, 2026
Period covered: Jul 14, 2026 to Oct 12, 2026
At a glance¶
-
2,710 users
Used Excel MCP during the last 90 days.
-
1,147,285 actions
Recorded across workbooks, cells, data, charts, and automation.
-
64.53% returned
Used Excel MCP on at least two different days.
Usage over the last 12 complete weeks¶
Each bar is one full week, which makes changes easier to compare.
Release upgrades over time¶
Each column is one week; the final column is the current week so far. A user appears once, under the latest release they used that week. This makes it easy to see newer releases replace older ones without overall user growth changing the scale. Less common releases are grouped as Other.
The latest two weeks¶
The latest period is Sep 28, 2026 to Oct 12, 2026. It is compared with Sep 14, 2026 to Sep 28, 2026.
What the numbers tell us¶
Summary written by GitHub Copilot
Copilot reads only the anonymous totals used to build this page. Its summary is checked automatically so it cannot add private details or numbers that are not in the report.
What changed¶
Over 90 days, 2,710 people used the tool and 64.53% came back more than once. Weekly users peaked at 830 in mid-August, fell to 323 in late September, then rose to 392. Yet estimated work kept climbing, from 185,182 to 632,495 a week. This suggests fewer, but busier, people.
In the last 14 days, users rose 15.63% while actions rose 90.77%. Actions growing faster than people may mean each person is doing more. People also moved to new releases quickly: 2.0.8 reached 56.28% of weekly users, and 2.3.9 already reached 27.59% in the newest partial week.
How people use it¶
Everyday cell and table work makes up 63.87% of actions, mostly reading and writing values. "Work" here is an estimate, not time or speed: quick reads count 1, everyday edits count 3 and heavy data work counts 10. By that measure, macros are 7.94% of actions but 16.97% of work, and Power Query data imports are 3.93% of actions but 10.72% of work. Advanced use is real: most people who tried macros (77.23%) or Power Query (71.66%) did so on their first day.
In the last 30 days, a middle AI assistant session had 9 actions, but sessions of 201 or more actions were 4.81% of sessions and 48.45% of actions. 66.03% of sessions mixed areas, most often cells with worksheets. Of new people, 51.56% returned after a week and 39.83% after three weeks. Workdays averaged 25,430 actions versus 12,876 on weekends. Typical waits ranged from 0.07 seconds for cell work to 2.26 seconds for Power Query; waits depend on workbook size and computer, so they do not score quality.
Since late September, AI assistant users (1,744) averaged 203.1 actions each, while command line users (145) averaged 538.9. These are different people over a shorter window, but command line users may lean more on cell work (80.44% of their actions).
Where people get it¶
GitHub release files lead with 24,115 downloads, then VS Code installs (8,875), the npm AI assistant package (4,856) and the npm command line package (3,213). Weekly npm downloads grew from 2,891 to 5,178. Release 2.3.3 (577) and 2.3.9 (435) were the most downloaded recent releases. Over 4 days to October 8, GitHub and VS Code gained 1,006 downloads.
Downloads count downloads, not people. One person can download several times, and updates and automated installs add to totals. Channels overlap, so totals should not be added. npm covers the last 12 months; others count since each channel started.
What we will improve¶
These are plans, not findings. We will make long sessions and heavy data work, like macros and Power Query, smoother. We will help new people find advanced features and keep returning after the first week.
What people use most¶
The bars group actions by the main features highlighted on the Excel MCP homepage. Each feature has two bars. Share of actions counts every action once. Share of work gives heavier actions more weight, so refreshing a Power Query counts for more than reading a few cells. See how share of work is calculated. Smaller capabilities are grouped as Other features.
Most common actions¶
How share of work is calculated¶
An estimate, not a measurement
Every action is given one of three fixed effort levels. Quick reads, such as reading cells or listing worksheets, count 1. Everyday edits, such as writing values or formatting cells, count 3. Heavy data work, such as refreshing Power Query or the Data Model, running macros, or building PivotTables, counts 10. The levels are chosen by the maintainers and checked automatically whenever an action is added. They describe the kind of work, not how long it took. See the full list of levels.
102,083 actions came from older releases that used action names which no longer exist. They are counted as actions but left out of share of work.
50.11% of people who used Excel MCP in this period did at least one piece of heavy data work.
Where most of the work goes¶
These actions add up to the largest share of estimated work. Common light actions can still appear here when they are used very often.
Command line and AI assistant¶
Excel MCP can be used through an AI assistant, which talks to the MCP Server, or directly from the command line with excelcli. Each action has recorded which of the two was used since Sep 28, 2026, so this comparison covers a shorter period than the rest of the page. The two groups are mostly different people doing different jobs, so differences describe how each is used, not which is better. A group is shown only when it has at least 10 users.
Share of people and actions¶
Someone who uses both is counted in each group.
What each group works on¶
Each feature's share of estimated work within each group.
Most common actions in each group¶
AI assistant (MCP Server)
Command line (excelcli)
How people work¶
These views use the last 30 days unless stated otherwise. Groups with fewer than 10 users are not shown.
Size of AI assistant sessions¶
A session is one run of the MCP Server inside an AI assistant. The command line is left out because every excelcli command runs on its own. Out of 14,623 sessions, the typical one had 9 actions, and 66.03% used two or more areas of Excel. Sessions with more than 200 actions were 4.81% of sessions but 48.45% of actions.
Areas used together¶
How often two areas of Excel appear in the same AI assistant session, as a share of all sessions.
Do new users come back?¶
Of 2,463 people first seen between 4 and 12 weeks ago, this is how many used Excel MCP again later. People first seen before the 90-day window may be counted as new.
Typical wait by area¶
How long a typical action takes from request to answer, including Excel's own work. The second number is the wait that 1 in 10 actions goes past. Waits depend on workbook size and the computer, so they show which areas are heavier, not how fast Excel MCP is.
Weekdays and weekends¶
Total actions on each day of the week, added up over the last 8 complete weeks (UTC). A single weekday averaged 25,430 actions, compared with 12,876 on a single weekend day.
When people first try advanced features¶
Among people who started using Excel MCP in the last 60 days, the share of each advanced area's users who first used it within a day of their first Excel MCP action.
Where people get Excel MCP¶
These numbers come from the public download counters on npm, GitHub, and the Visual Studio Marketplace, checked on Oct 12, 2026.
Downloads are not people
One person can download Excel MCP many times, updates and automatic installs add to the counts, and the same person can appear in more than one place. Use these numbers to compare channels and spot trends, not to count users, and do not add them together.
Downloads by channel¶
npm counts cover the last 12 months. The other counts include everything since each channel started.
npm downloads each week¶
npm keeps a daily history, so each bar is one full week of downloads for the MCP Server and command line packages together.
Downloads of recent releases¶
Files downloaded from each GitHub release: the MCP Server and command line packages, the VS Code extension files, and the Claude Desktop bundle. Newer releases have had less time to collect downloads.
New downloads each week¶
GitHub releases and the VS Code Marketplace only publish running totals. This chart shows how much those totals grew between one weekly report and the next. Bars marked * cover a gap between reports that was not one week and show the average for one week of that gap.
How this report protects privacy¶
The report is built from anonymous counts and percentages. We do not publish or give Copilot user or session codes, file fingerprints, locations, messages, workbook content, error messages, or technical error details.
Excel MCP never intentionally collects workbook contents, cell values, formulas, prompts, messages, file names or paths, names, email addresses, or account details. Read the full privacy policy.
You can inspect exactly how the report is built in Update-UsageAnalytics.ps1 and usage-analytics.yml.