Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchPC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11One developer’s Claude Code logs show a sharp difference between the time an individual MCP server takes to connect and the time a session spends waiting for all configured servers. Across 27,257 clean connection measurements, the median was 582 ms and the 90th percentile (p90) was 3,678 ms. But the p90 for total connection time per session was 35.5 seconds.
Those numbers come from Achia Cohen’s analysis of one machine and one client—not a representative benchmark of MCP users. They are most useful as a case study in why per-server latency and the combined wait across a session answer different questions.
As an Amazon Associate I earn from qualifying purchases.
What the 27,257-connection report measured
Cohen analyzed Claude Code connection logs spanning 35 days, from August 13 to September 16, 2026. The report counts 33,599 log files, 38,876 connection attempts, 22 MCP servers, and 2,888 sessions. These are the author’s counts from one machine and one client, not population-level measurements.
For the latency distribution, the author paired log lines marking a connection start and the later establishment of server capabilities. That yielded 27,257 clean start-to-established pairs. The report gives a median connection latency of 582 ms, a p90 of 3,678 ms, and a 99th percentile (p99) of 14,049 ms.
#1 Best Overall
The author capped elapsed measurements at the client’s stated 30,000 ms timeout to avoid counting laptop sleep as active connection time. In a cleanup of 27,281 samples, 24 were removed. The cap is important when interpreting the long tail: this is a log-based measurement with a defined ceiling, not an unconstrained measure of how long every possible attempt might take.
Why a session’s p90 wait was 35.5 seconds
The 35.5-second p90 is the report’s total connection time per session, not the p90 latency of one server. The median total per session was 5.9 seconds. Cohen reports a median of eight MCP servers per session, so even connections that are individually quick can add up to a noticeable wait when several servers initialize in the same session.
These figures describe different units: milliseconds per clean connection pair versus total connection time per session. They should not be read as contradictory, nor should the session total be treated as the delay of each server. The report’s session-level result is the more relevant figure for a user asking how long the combined startup wait can feel; the per-connection percentiles help show how that total may be distributed.
Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Fix the driver behind crashes, sound loss and screen glitches3Clear out junk files and repair common Windows errorsWhat the transport and token-refresh comparisons show
Within this dataset, HTTP connections had a lower p90 than stdio connections, while stdio had a lower median. The numbers describe this sample only and do not establish a generally faster transport.
Rank #3
| Connection grouping | Median | p90 | What the report supports |
|---|---|---|---|
| HTTP | 676 ms | 3,395 ms | Higher median but lower p90 than stdio in this dataset. |
| stdio | 222 ms | 4,693 ms | Lower median but higher p90 than HTTP in this dataset. |
Transport choice also depends on deployment context—for example, whether an HTTP service is already running or a stdio process needs to be spawned. The measurements do not isolate those conditions or prove which transport will be faster in another setup.
The report also found longer observed connection times when a token refresh occurred during the connection window:
Rank #4
- Server 2022 Standard 16 Core
| Connection grouping | Median | p90 |
|---|---|---|
| Token refresh during the connection window | 3,170 ms | 9,619 ms |
| No refresh during the connection window | 550 ms | 3,400 ms |
This is an association in one dataset, not evidence that token refresh alone caused the delay. Other conditions could also differ between the groups.
Quick wins for a faster PC:
Repair Windows errors before they cause bigger problemsFix Now →Scan for outdated or missing drivers - takes under a minuteDriver Scan →How to interpret attempts without an established log line
Of 38,876 recorded attempts, 11,556 (29.7%) did not have a corresponding logged established line. Cohen cautions that log rotation may account for some missing records. The figure is therefore an upper-bound signal for attempts without a recorded establishment, not a verified connection-failure rate.
A missing line in the logs cannot, by itself, tell you whether a server failed, whether the relevant record was rotated out, or whether another logging condition intervened. Treating every missing establishment record as a failed connection would overstate what the report establishes.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What you can check in your own MCP setup
The report’s practical suggestions focus on configuration and existing logs rather than on treating its results as a universal benchmark:
- Count configured servers. A session that initializes several servers can have a substantial combined wait even when many individual connections are quick.
- Make heavy servers opt-in where appropriate. If a server is not needed in every session, reducing the set initialized by default may reduce unnecessary startup work.
- Look for timing patterns around token refresh. The report identifies an association worth checking in your own logs, but it does not establish refresh as the cause.
- Compare transports in context. Examine both median and tail latency, and account for whether a service is already running or a process must be started.
- Inspect logs already available to your client. Cohen describes logs under
~/Library/Caches/claude-cli-nodejs/<project>/mcp-logs-<server>/for the setup analyzed. That location is specific to the described environment; verify the log location for your installed client version rather than assuming the same path applies everywhere.
What the report can—and cannot—tell you
Cohen’s framing is deliberately personal: “This is one developer’s machine, not a lab. That’s the point — it’s the distribution you actually live in.” That makes the report a useful view of real-world logs on one setup, but it does not show that other developers, machines, clients, or server configurations have the same distribution.
Free tools Windows power users keep installed
One-click scans. No signup required.
It cannot establish typical MCP latency across users, prove that OAuth refresh causes slower connections, or identify a universally optimal transport. Cohen explicitly asks readers whether a median of eight configured servers is normal or an outlier, underscoring that broader comparisons are still needed.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
