Static HTML Export

TalkToFigma Desktop GA4 Analysis

The data shows a desktop product that did more than launch. Users connected it, ran real Figma MCP tool calls, and exposed a clear product lesson: the core value was reducing anxiety between first launch and first connection.

Created: 2026-07-01 Analysis period: 2025-07-01 to 2026-06-30 Effective data start: 2025-08 Source: GA4 Figma MCP Agent / properties/499789310

Presentation-Ready Metrics

These are the safest top-line numbers to use in a retrospective. Total events are included for telemetry scale, but the interpretation should stay cautious because exception volume is large.

Active users

31,360

Meaningful scale for a desktop utility over one year.

MCP tool calls

570,373

Strongest evidence of actual workflow usage.

Success rate

94.1%

Among MCP calls with explicit success or failure values.

Country values

165

Includes '(not set)', so use as spread evidence carefully.

Metric	Value	Interpretation
Active users	31,360	Meaningful one-year scale for a desktop utility.
Sessions	24,429	Evidence of repeated use, while desktop telemetry may understate web-style sessions.
Total events	1,826,813	Telemetry scale, not a pure success metric because exceptions inflate the count.
MCP tool calls	570,373	Best evidence of workflow usage beyond simply launching the app.
Successful MCP calls	536,267	About 94.1% success among explicit success or failure outcomes.
Country values	165	Includes '(not set)', so treat as global spread evidence rather than exact countries.
Top country	South Korea, 4,263 active users	Largest active-user country in the snapshot.
Desktop share	100% desktop	This dataset describes desktop-app telemetry, not a web funnel.

Narrative

One-line takeaway: In the end, the core value was not adding more features. It was reducing the anxiety between first launch and first connection.

This project was less about adding one more feature to Figma and more about creating an installable desktop path into the Figma MCP workflow. The app reached 31,360 active users and recorded 570,373 MCP tool-call events. Among calls with explicit success or failure values, 536,267 succeeded and 33,777 failed.

The top MCP actions look like real design-editing work: creating text, setting fill colors, creating frames, adjusting corner radius, editing text content, creating rectangles, inspecting nodes, and exporting images. That supports the claim that users did not merely open the app. They used it to perform actual work through MCP.

The retrospective should not frame the project as a failure. A better frame is that the project proved a need, then the need moved toward larger platform surfaces such as official Figma MCP and Canvas-style agents. The hard part is accepting that a working independent tool can become lower-priority precisely because the core direction was validated.

Growth And Usage Depth

Monthly Active Users Active users rose quickly after launch and peaked in January 2026.

Monthly Event Volume March 2026 has the event-count peak, but the spike includes heavy exception volume.

Month	Active users	Sessions	Events
2025-08	243	185	1,502
2025-09	311	195	2,037
2025-10	474	292	4,366
2025-11	1,150	660	8,351
2025-12	3,323	1,981	14,398
2026-01	7,890	4,697	44,976
2026-02	5,410	3,429	107,221
2026-03	6,084	4,738	826,851
2026-04	3,153	3,261	311,206
2026-05	2,602	2,636	211,318
2026-06	2,505	2,576	294,587

MCP Work Happened

The tool-call distribution is the strongest evidence that the app enabled real Figma workflows rather than only app starts or page views.

Top Successful MCP Tool Calls Successful events by tool name.

MCP Call Outcome 536,267 successful calls and 33,777 failed calls among explicit outcomes.

Success 94.1% Failure 5.9%

MCP tool	Successful events	Active users
create_text	105,643	613
set_fill_color	92,561	604
create_frame	52,509	613
set_corner_radius	38,053	504
set_text_content	37,845	455
create_rectangle	34,779	473
get_node_info	32,736	842
move_node	13,859	324
export_node_as_image	13,536	455
set_stroke_color	10,910	361

Global Reach And Platform Shape

GA4 recorded 165 country values and 37 language values. Because country includes '(not set)', the safer claim is broad global spread rather than an exact country count.

Top Countries By Active Users Excludes '(not set)' for presentation clarity.

Operating System Active Users Windows drove breadth; MacOS generated deeper MCP call volume.

Country Table

Country	Active users	Sessions	Events
South Korea	4,263	5,599	1,143,307
Russia	3,133	1,969	132,773
India	2,774	1,839	25,859
United States	2,047	1,580	132,057
Nigeria	1,824	1,095	9,811
Japan	801	565	5,070
Indonesia	591	369	3,678
Brazil	565	405	12,648
Singapore	564	461	77,761

Language Table

Language	Active users	Sessions	Events
English	18,875	15,497	1,485,167
Russian	4,094	2,489	21,686
Korean	2,766	2,759	264,062
French	1,209	744	5,224
Chinese	990	620	31,502

First Launch And First Connection

This should be the center of the retrospective. The app's value was not only 'let AI control Figma'. It made the connection server visible, recoverable, and understandable.

First-Connection Journey The strongest product surface was launch, server state, tutorial, OAuth, and configuration.

App start

30,912

Start WebSocket server

21,115

Tutorial shown

6,875

Tutorial completed

4,126

Direct OAuth completed

1,620

Copy MCP config

1,495

Connection Flow Screens Page titles confirm an explicit setup journey and diagnosis surface.

Event or action	Active users	Events
app_start	30,912	40,514
start_websocket_server	21,115	23,898
direct_oauth_start	3,016	4,382
copy_mcp_config	1,495	2,448
report_issue	99	133
Direct OAuth completed	1,620	2,124
kill_all_servers	540	1,017

Operational Burden

The app worked, but it also behaved like a real desktop product with a broad failure surface: port conflicts, server restarts, OAuth, WebSocket state, logs, and network issues.

Exception Messages By Affected Users Affected user count is safer than raw repeated event volume for operational prioritization.

Custom Error Messages By Affected Users Timeout and port availability issues reached many users.

Error message	Events	Active users
write EPIPE	897,378	12
net::ERR_TUNNEL_CONNECTION_FAILED	99,996	1
write EIO	56,708	1
EPIPE: broken pipe, write	20,036	1
Attempted to register a second handler for 'server:start'	2,605	1,231
Address already in use: bind	1,539	1,539
Address already in use	756	552
Timed out waiting for 30000 ms	650	102
Not connected to Figma	334	32
Must join a channel before sending commands	268	148

A key caveat: write EPIPE produced 897,378 events from only 12 active users. The operational story should therefore combine raw volume with affected user count.

KPT Candidates

Keep

Focused on reducing anxiety from first launch to first connection.
Turned the Figma MCP workflow into an installable desktop experience.
Instrumented enough analytics to verify usage scale, geography, OS distribution, MCP success, onboarding behavior, and operational burden.
Built a product, not just a test environment.

Problem

Once the core capability is absorbed by official platform features, maintaining the same update intensity becomes hard to justify.
Exception events accumulated heavily and made some event-count metrics hard to interpret.
Connection, port, server state, OAuth, and logs became a larger product surface than expected.

Try / Learning

Define the user context that platforms are less likely to absorb, not only the feature.
Separate first successful connection, reconnection, config copy, success, failure, and diagnostic actions more rigorously.
Track affected active users alongside raw error-event counts.
Decide future major updates by combining usage, platform absorption risk, operational burden, and personal sustainability.

Source Details And Caveats

This export is a standalone static HTML artifact. It does not include the MCP app shell, interactive top bar, refresh controls, export controls, or app-only navigation.

Field	Value
Original artifact URL	https://mcp-server-dataanalyticswidgets-d90d6b74b2c37858.web-sandbox.oaiusercontent.com/?app=skybridge
Artifact generated at	2026-07-01T00:00:00+09:00
Static export file	`talk-to-figma-desktop-ga4-static-report.html`
GA4 source	GA4 Analytics MCP
Property	Figma MCP Agent / `properties/499789310`
Requested range	2025-07-01 to 2026-06-30
Observed effective range	2025-08 to 2026-06
Timezone	Asia/Seoul
Local source snapshots	`docs/analytics/ga4/talk-to-figma-desktop-ga4-analysis.en.md`, `data/ga4/talk-to-figma-desktop-ga4-raw-data.json`, `data/ga4/talk-to-figma-desktop-ga4-raw-data.csv`, `data/ga4/talk-to-figma-desktop-ga4-raw-data.md`

Country row_count is 165, but this JSON stores the top rows used in the analysis snapshot. See the Markdown file for more displayed country rows.
Region and city reports were dominated by '(not set)' values and are not reliable for fine-grained location claims.
Browser, customEvent:platform, and customEvent:os_info are mostly '(not set)' and should not be used as strong segmentation evidence.
Exception event counts are inflated by repeated messages from very small numbers of users, especially write EPIPE.
GA4 funnel report semantics are directional and ordered; funnel counts do not equal broad event totals.
Total event count includes 1,081,199 app_exception events, so it should not be used as a pure usage metric.
GA4 funnel numbers are ordered funnel results and should not be directly compared with broad event totals.
averageSessionDuration looks unusually high in this desktop telemetry context and is intentionally not used as a primary claim.