[Bug] Jira Authentication Fails: Persistent Reconnection and Session Issues

NOTE: my claude.ai IS CONNECTED to jira MCP server just claude code I cannot get to connect.

I cannot tell if this is atlassian bug or claude code bug since claude.ai works with jira MCP server but claude code does not.
Here is a post on their side -> https://community.atlassian.com/forums/Jira-questions/claude-code-with-jira-MCP-constant-fail/qaq-p/3125446#M1151925

I can confirm this is a regression in version 2.0.10. The Atlassian MCP server works perfectly in 2.0.9 but fails in 2.0.10.

Root cause: Version 2.0.10 removed SSE (Server-Sent Events) support from native builds. The Atlassian MCP server endpoint is https://mcp.atlassian.com/v1/sse which requires SSE protocol.

Debug log evidence from 2.0.10:

[DEBUG] MCP server "atlassian": Connection failed after 0ms: SSE MCP servers are currently not supported in native builds.
[ERROR] MCP server "atlassian" Connection failed: SSE MCP servers are currently not supported in native builds.

Authentication succeeds (OAuth tokens are valid), but the connection immediately fails due to protocol incompatibility.

Workaround:
Downgrade to version 2.0.9.

Then restart Claude Code. The Atlassian MCP will connect successfully.

Timeline:

• Version 2.0.9 (released Oct 7): Works ✅

<img width="1696" height="824" alt="Image" src="https://github.com/user-attachments/assets/dc23ca39-f6cf-4985-a216-8568af3e76dd" />

• Version 2.0.10 (released Oct 8): Broken ❌

<img width="1712" height="974" alt="Image" src="https://github.com/user-attachments/assets/12e49ea3-9bb2-4f9e-a53e-18f73ca08f98" />

This appears to be an unintended breaking change in 2.0.10. SSE support needs to be restored for Atlassian MCP to function.

---
This investigation was conducted with assistance from Claude Code.

Odd, I reverted hoping it would work even clearing authentication but mine keeps breaking (see version of 2.0.9 is there now after my revert but still fails)

<img width="1322" height="307" alt="Image" src="https://github.com/user-attachments/assets/5cac9a01-fad1-4992-a9d0-55cada6be911" />

Man, how did you get to those logs. I ask claude and it scans my filesystem without finding them. I got this but my /help does not show this toggle dev tools >

1. Opening Claude Code
2. Going to Help → Toggle Developer Tools
3. Checking the Console tab for the MCP error messages like the DEBUG/ERROR lines you mentioned

The application state and settings are in ~/.claude/ but the runtime logs are in the developer console, not written to disk.

I am getting exactly same issue

> /mcp
  ⎿  Authentication successful, but server reconnection failed. You may need to manually restart Claude Code for the changes to take effect.

Working again after downgrading to 2.0.9

Man, how did you get to those logs. I ask claude and it scans my filesystem without finding them. I got this but my /help does not show this toggle dev tools > 1. Opening Claude Code 2. Going to Help → Toggle Developer Tools 3. Checking the Console tab for the MCP error messages like the DEBUG/ERROR lines you mentioned The application state and settings are in ~/.claude/ but the runtime logs are in the developer console, not written to disk.

claude said I should run with claude --debug. And it logs in ~/.claude/debug/latest. I asked claude to look at it after I restarted claude and resumed the conversation. 😂

Getting this error with the Atlassian MCP server on Claude 2.0.10:

[DEBUG] MCP server "atlassian": Connection failed after 0ms: SSE MCP servers are currently not supported in native builds.
[ERROR] MCP server "atlassian" Connection failed: SSE MCP servers are currently not supported in native builds.

I can confirm rolling back to 2.0.9 works

Can confirm this issue affects multiple Atlassian MCP server instances on macOS Darwin 26.

Tested and working on 2.0.9:

• Multiple Confluence instances (various read/write operations)
• Jira instance (issue management, search)

All authenticate successfully via OAuth and retrieve data without issues on 2.0.9.

Workaround to prevent auto-update:

Add to ~/.zshrc (or your shell rc file):

# Claude Code: Disable auto-updates (pinned to v2.0.9 due to Atlassian MCP SSE bug)
export DISABLE_AUTOUPDATER=1

Then restart your shell and downgrade to 2.0.9.

The SSE removal in 2.0.10 is a breaking change for the Atlassian MCP endpoint (https://mcp.atlassian.com/v1/sse). This affects both Confluence and Jira integrations.

FWIW I also confirmed this is broken in 2.0.10 and 2.0.11 but not .9.

It's not only Jira that's broken, but also Linear's MCP for me.

ok, this is odd as sentry is working ->

"atlassian": {
"type": "http",
"url": "https://mcp.atlassian.com/v1/sse"
},
"sentry": {
"type": "http",
"url": "https://mcp.sentry.dev/mcp"
}

Because the transport type is http @deanbiltup

@deanbiltup this works in claude CLI, but just know that the VS code extension in MacOSX doesn't work with it:

  "mcpServers": {
    "atlassian": {
      "type": "sse",
      "url": "https://mcp.atlassian.com/v1/sse"
    }
  },

The VS code extension fails to read the authentication information (seems to look to a different place than the CLI for the data). So the MCP server will not work in the VS code extension.

CLI version 2.0.12 (Claude Code)
VS code extension

Name: Claude Code for VS Code
Id: Anthropic.claude-code
Description: Claude Code for VS Code: Harness the power of Claude Code without leaving your IDE
Version: 2.0.12
Publisher: Anthropic
VS Marketplace Link: https://marketplace.visualstudio.com/items?itemName=anthropic.claude-code

Confirming this issue - blocking production workflow

Environment:

• Platform: macOS (darwin)
• Claude Code: Latest version
• Atlassian Instance: Jira Cloud (caylent.atlassian.net)
• Use Case: Active development project requiring frequent JIRA ticket updates

Exact Symptom:
Every single authentication attempt results in:

Authentication successful, but server reconnection failed. You may need to manually restart Claude Code for the changes to take effect.

After restarting Claude Code as instructed:

• Authentication is lost
• MCP shows as not connected
• Must authenticate again, which fails with same error
• This creates an infinite loop making the Atlassian MCP completely unusable

Impact:
This is blocking our development workflow where we need to:

• Query JIRA tickets programmatically via Claude Code
• Update ticket status and fields using editJiraIssue
• Automate JIRA interactions as part of our development process

What Works:

• Initial OAuth flow completes successfully
• Browser authentication page works
• Tokens appear to be generated

What Fails:

• Server reconnection after authentication
• Persistence of authentication state
• Any actual use of Atlassian MCP tools

Workarounds Attempted:

• Multiple restarts of Claude Code ❌
• Re-authentication ❌
• Clearing authentication and starting fresh ❌
• None successful - issue persists 100% of the time

This is the only MCP server giving us problems - other MCP servers work correctly. The Atlassian/JIRA MCP is critical for our workflow and this bug makes it completely non-functional.

Would greatly appreciate any updates on this issue or suggested workarounds. Happy to provide additional diagnostics if helpful.

if anyone needs a workaround for the time being, this worked for me

claude mcp remove atlassian
claude mcp add --scope user --transport stdio atlassian -- npx mcp-remote@0.1.13 https://mcp.atlassian.com/v1/sse

Harry's workaround actually works. Thank you Harry.

SSE MCP servers are currently not supported in native builds.

I'm also experiencing this with multiple servers that we use. This seems like a pretty big breaking change, that I don't see mentioned in the docs literally anywhere. Can it be confirmed from someone on the anthropic team that this is the case moving forward, and maybe some information as to why?

@harry-da amazing! thank you!

if anyone needs a workaround for the time being, this worked for me `` claude mcp remove atlassian claude mcp add --scope user --transport stdio atlassian -- npx mcp-remote@0.1.13 https://mcp.atlassian.com/v1/sse ``

And, how did that work actually? What is the logic behind? It worked for me as well, but I just wanted to know the logic behind it. Thank you.

Very rough explanation: It uses an intermediary MCP server called 'mcp-remote' to translate atlassian's SSE server into STDIO, which Claude Code knows how to use.

I'd say this is more of a bug for Atlassian's MCP than Claude Code, right? Given SSE is deprecated per https://github.com/modelcontextprotocol/modelcontextprotocol/pull/206.

Deprecated, but still widely used and present in many examples Claude Code's MCP docs. There's likely a case to remove it, but not without warning.

@arobinson I modified {repo}/.mcp.json to (http to see keeping sentry to http as that works) AND restart claude, /mcp and clear authentationc and authenticate and NOT WORK :( ->

    "atlassian": {
      "type": "sse",
      "url": "https://mcp.atlassian.com/v1/sse"
    },
    "sentry": {
      "type": "http",
      "url": "https://mcp.sentry.dev/mcp"
    },

error ->

/mcp

⎿  Authentication successful, but server reconnection failed. You may need to manually restart Claude Code for the changes to take effect.

Version now is 2.0.14 ( I upgraded again ).

@Ubeydu wow - thanks! WORKAROUND works!!!

OUCH - @Ubeydu that workaround keeps failing and I have to restart claude all the time and at very inconvenient times. any ideas how to fix that on mac? perhaps it is because I leave and come back an hour later or something.

ANYONE - upvote the atlassian one to have this fixed -> https://github.com/atlassian/atlassian-mcp-server/issues/8

OUCH - @Ubeydu that workaround keeps failing and I have to restart claude all the time and at very inconvenient times. any ideas how to fix that on mac? perhaps it is because I leave and come back an hour later or something.

Same for me. The workaround sometimes works, but often randomly fails.

It's not only Jira that's broken, but also Linear's MCP for me.

The fix for Atlassian works for Linear too:
claude mcp remove linear-server

claude mcp add --scope user --transport stdio linear-server -- npx mcp-remote@0.1.13 https://mcp.linear.app/sse

I started getting this error today while authenticating on atlassian redirection
upstream connect error or disconnect/reset before headers. reset reason: local connection failure

This is still broken for me as of version 2.0.28.

This issue has been inactive for 30 days. If the issue is still occurring, please comment to let us know. Otherwise, this issue will be automatically closed in 30 days for housekeeping purposes.

Still happening today with latest claude 2.0.64. Workaround didn't work/no longer works - just get Authentication successful, but server reconnection failed. You may need to manually restart Claude Code for the changes to take effect

v2.0.67 here. claude mcp add --scope user --transport sse atlassian https://mcp.atlassian.com/v1/sse is working for me.

This issue has been automatically closed due to 60 days of inactivity. If you're still experiencing this issue, please open a new issue with updated information.

This issue has been automatically locked since it was closed and has not had any activity for 7 days. If you're experiencing a similar issue, please file a new issue and reference this one if it's relevant.