# Windows mcp.json major bug with non-stdio servers

**URL:** <https://forum.cursor.com/t/windows-mcp-json-major-bug-with-non-stdio-servers/145414>\
**Category:** Bug Reports\
**Created:** [December 6, 2025, 10:41am UTC](https://forum.cursor.com/t/windows-mcp-json-major-bug-with-non-stdio-servers/145414 "2025-12-06T10:41:55Z")\
**Posts on this page:** 8\
**Page:** 1

<div class="post-metadata">

**Author:** ![Chris\_Drake](https://avatars.discourse-cdn.com/v4/letter/c/da6949/32.png) [@Chris\_Drake](https://forum.cursor.com/u/Chris_Drake)\
**Post date:** [December 6, 2025, 10:41am UTC](https://forum.cursor.com/t/windows-mcp-json-major-bug-with-non-stdio-servers/145414/1 "2025-12-06T10:41:55Z")

</div>

### Where does the bug appear (feature/product)?

Cursor IDE

### Describe the Bug

Cursor establishes N duplicate connections to every mcp server listed in mcp.json where “N” is the number of servers in that file.

e.g. if you have 5 servers, cursor establishes 5 connections to each on of those 5 servers - 25 in total!

This bug has been around for ages - PLEASE FIX IT.

### Steps to Reproduce

add 2 more more non-stdio mcp servers

### Expected Behavior

1 connection per server only.

### Operating System

Windows 10/11

### Current Cursor Version (Menu → About Cursor → Copy)

Version: 2.1.49 (system setup)  
VSCode Version: 1.105.1  
Commit: 21a2ed198584d56a91c0b996d1a09c93f8538440  
Date: 2025-12-05T07:49:06.016Z  
Electron: 37.7.0  
Chromium: 138.0.7204.251  
Node.js: 22.20.0  
V8: 13.8.258.32-electron.0  
OS: Windows\_NT x64 10.0.17763

### Additional Information

This is a massive problem for infrastructure operators operating public mcp servers - we have to handle at least 2 x N connections per client (backchannel plus requests) - so we’re having single users with 100 or more continuous connections to our servers because of this bug.

### Does this stop you from using Cursor

No - Cursor works, but with this issue

---

<div class="post-metadata">

**Author:** ![deanrie](https://sea3.discourse-cdn.com/cursor1/user_avatar/forum.cursor.com/deanrie/32/1263_2.png) [@deanrie](https://forum.cursor.com/u/deanrie)\
**Post date:** [December 6, 2025, 3:29pm UTC](https://forum.cursor.com/t/windows-mcp-json-major-bug-with-non-stdio-servers/145414/2 "2025-12-06T15:29:22Z")

</div>

Hey, thanks for the report. Confirming a known issue on Windows with duplicated connections to non-stdio MCP servers (N×N instead of N). Related threads:

- [Cursor MCP server subsystem is totally SNAFU - Please can you prioritize some fixes!](https://forum.cursor.com/t/cursor-mcp-server-subsystem-is-totally-snafu-please-can-you-prioritize-some-fixes/143156)
- [Duplicated log lines when connecting to MCP servers](https://forum.cursor.com/t/duplicated-log-lines-when-connecting-to-mcp-servers/137761)

This bug is critical for public MCP server operators - you may see N^2 client connections.

Could you share details for escalation:

- mcp.json contents (global `~/.cursor/mcp.json` and project `.cursor/mcp.json` if both exist)
- Server types and endpoint URLs (SSE/HTTP)
- Server-side logs showing duplicated connections
- Logs from Help \> Toggle Developer Tools \> Console when MCP starts
- Whether it reproduces with one non-stdio server (no dupes) vs two+ (duplication appears)
- Whether you tested earlier Cursor versions

Temporary workarounds:

- Ensure a server isn’t defined in both `~/.cursor/mcp.json` and `.cursor/mcp.json`
- Prefer stdio over SSE where possible
- Use a single Cursor window
- Disable extra MCPs in Settings \> MCP

I’ll forward this to the team with your details.

---

<div class="post-metadata">

**Author:** ![Chris\_Drake](https://avatars.discourse-cdn.com/v4/letter/c/da6949/32.png) [@Chris\_Drake](https://forum.cursor.com/u/Chris_Drake)\
**Post date:** [December 6, 2025, 9:20pm UTC](https://forum.cursor.com/t/windows-mcp-json-major-bug-with-non-stdio-servers/145414/3 "2025-12-06T21:20:48Z")

</div>

Hi Dean,

With respect - I am your paying customer, and this bug has been reported many times already.

It looks like it’s never made it past your team (?triage area?) into “engineering” in the first place?

You are asking me to do significant research for you, with no guarantee that you’ll even send the results to your team, no guarantee it will be fixed, and no compensation to me for my work.

Don’t get me wrong - we REALLY want this bug fixed - but it only takes \< 1 minute for someone in your engineering team to verify this, which they really should be doing themselves anyhow, (try [GitHub - open-webui/mcpo: A simple, secure MCP-to-OpenAPI proxy server](https://github.com/open-webui/mcpo) for an easy test - you can see the mistakes on-screen) ESPECIALLY is they should be accumulating tests for these bugs so you can do basic quality-assurance before you keep shipping new product versions:

> [@Stop Shipping Unstable Builds](https://forum.cursor.com/t/stop-shipping-unstable-builds/143002):
>
> I’ve been using Cursor since the 0.x versions. In the past, I was genuinely excited about every new release — I would literally click Check for Updates almost every day. Lately, however, I feel the opposite: I hesitate whenever I see a new version available. The reason is simple: almost every update breaks something. If it’s not one issue, it’s another. Version 1.7 felt relatively solid and stable. But from 2.0 onwards, things have gone downhill — and there are already many posts out there abo…

Please let me know what’s going to happen, and a guarantee that it will be sent to “engineering”, when I expend the effort to collect all the information you’re asking for here. Can you commit to providing me with a ticket-number so we can track “engineering progress” towards fixing this bug after I file this info for you?

As a customer, and a programmer, I feel it is my duty to URGE you and your team to put some kind of quality-assurance system between whoever hits “build” and whoever hits “publish” - the staggering amount of new bugs and broken things that show up daily in windows builds proves beyond all doubt that you have no functional quality system at Cursor - please can you ask your team to put something in!! It’s really easy to automate that - Claude is excellent at testing itself inside Cursor - _WE_, your customers, should **not** be the only people doing that!!!

---

<div class="post-metadata">

**Author:** ![Chris\_Drake](https://avatars.discourse-cdn.com/v4/letter/c/da6949/32.png) [@Chris\_Drake](https://forum.cursor.com/u/Chris_Drake)\
**Post date:** [December 9, 2025, 9:45pm UTC](https://forum.cursor.com/t/windows-mcp-json-major-bug-with-non-stdio-servers/145414/5 "2025-12-09T21:45:52Z")

</div>

This bug is still present in Version 2.2.1 (it established 10 connections to my 1 server at startup).

---

<div class="post-metadata">

**Author:** ![Chris\_Drake](https://avatars.discourse-cdn.com/v4/letter/c/da6949/32.png) [@Chris\_Drake](https://forum.cursor.com/u/Chris_Drake)\
**Post date:** [December 18, 2025, 9:32pm UTC](https://forum.cursor.com/t/windows-mcp-json-major-bug-with-non-stdio-servers/145414/6 "2025-12-18T21:32:59Z")

</div>

Still present in 2.2.36

---

<div class="post-metadata">

**Author:** ![Chris\_Drake](https://avatars.discourse-cdn.com/v4/letter/c/da6949/32.png) [@Chris\_Drake](https://forum.cursor.com/u/Chris_Drake)\
**Post date:** [December 24, 2025, 1:00am UTC](https://forum.cursor.com/t/windows-mcp-json-major-bug-with-non-stdio-servers/145414/7 "2025-12-24T01:00:53Z")

</div>

Still present in 2.2.43

See also: [New MCP bugs on non-stdio MCP servers - #4 by Chris\_Drake](https://forum.cursor.com/t/new-mcp-bugs-on-non-stdio-mcp-servers/146819/4)

---

<div class="post-metadata">

**Author:** ![Chris\_Drake](https://avatars.discourse-cdn.com/v4/letter/c/da6949/32.png) [@Chris\_Drake](https://forum.cursor.com/u/Chris_Drake)\
**Post date:** [January 3, 2026, 1:44pm UTC](https://forum.cursor.com/t/windows-mcp-json-major-bug-with-non-stdio-servers/145414/8 "2026-01-03T13:44:57Z")

</div>

Most of these bugs still present in:

Version: 2.3.21 (system setup)  
VSCode Version: 1.105.1  
Commit: 68e0a0385b87408d050869ea543e3778ad53f780  
Date: 2026-01-02T23:46:13.381Z  
Electron: 37.7.0  
Chromium: 138.0.7204.251  
Node.js: 22.20.0  
V8: 13.8.258.32-electron.0  
OS: Windows\_NT x64 10.0.17763

It had over 500 connections to my mcp server !!!

The bug with it calling the wrong server when tools have same names seems fixed now at least.

---

<div class="post-metadata">

**Author:** ![system](https://sea3.discourse-cdn.com/cursor1/user_avatar/forum.cursor.com/system/32/99069_2.png) [@system](https://forum.cursor.com/u/system)\
**Post date:** [January 25, 2026, 1:45pm UTC](https://forum.cursor.com/t/windows-mcp-json-major-bug-with-non-stdio-servers/145414/9 "2026-01-25T13:45:14Z")

</div>

Closed on January 25, 2026, 1:45pm UTC

This topic was automatically closed 22 days after the last reply. New replies are no longer allowed.
