MCP OAuth has stopped working a few days ago

[Andrei Kondratev]

Today I noticed that OAuth roundtrip has stopped working for me,

I am connecting from a local command line tool, which has been working fine recently (a day or two ago).

I tried to debug a little bit, what I found is that all

`https://mcp.atlassian.com/.well-known/oauth-protected-resource/v1/sse`

`https://mcp.atlassian.com/.well-known/oauth-protected-resource`

and 

`https://mcp.atlassian.com/register`

all returning 404

The client metadata sent to `/register` is

```

{
"redirect_uris": [
"http://127.0.0.1:59271/oauth-callback"
],
"client_name": "Gaunt Sloth Assistant",
"client_uri": "https://github.com/Galvanized-Pukeko/gaunt-sloth-assistant",
"software_id": "1dd38b83-946b-4631-8855-66ee467bfd68",
"scope": "mcp:read mcp:write",
"token_endpoint_auth_method": "none",
"grant_types": [
"authorization_code",
"refresh_token"
],
"response_types": [
"code"
]
}

```

I understand this may be due to recent changes in "available urls", but what would be a URL in this context? client_uri? Is it relevant in the context of local application?

I will appreciate any help.

[Ibraheem Osama]

Hey Andrei

Can you try this URL instead `https://mcp.atlassian.com/v1/register` for client registration?

Plus the MCP Server is still on sse so the well know is under https://mcp.atlassian.com/.well-known/oauth-authorization-server

Please let me know how it goes 

Thanks,
Ibby

[Andrei Kondratev]

I modified client url to point to the website rather than to the repo and added a domain in settings, but still getting 404 for /register

[Andrei Kondratev]

Thank you @Ibraheem Osama ! The v1 returned 201 and I went a little bit further!

now this one returns 404

https://mcp.atlassian.com/authorize?response_type=code&client_id=SOME_HASH&code_challenge=SOME_HASH&code_challenge_method=S256&redirect_uri=http%3A%2F%2F127.0.0.1%3A59786%2Foauth-callback&state=SOME_HASH&scope=mcp%3Aread+mcp%3Awrite

I guess this one also needs v1

I wonder what has changed? In the latest app I'm using https://github.com/modelcontextprotocol/typescript-sdk/releases/tag/1.21.1 to deal with OAuth

[Andrei Kondratev]

tried adding v1 to authorize ... and it gives "internal server error" 

[Andrei Kondratev]

There are two issues, it is either a bug in 

@modelcontextprotocol/sdk 1.21.1

or some incompatibility of jira and mcp sdk

The `internal server error` of /v1/authrorize happened because the client_id was already used, so I have removed the cached client metadata and authroize worked.

@Ibraheem Osama can you advise if I should clear cached metadata with client_id every time I log in?

My symptomatic treatment which worked:

1. Pinned @modelcontextprotocol/sdk to 1.21.0

2. Removed client information cache (stuff with client_uri, client_id, client_id_issued_at)