modelcontextprotocol / modelcontextprotocol/python-sdk
Trailing slash in OAuthMetadata's `issuer` causes issues with clients
まだ誰も着手していません。
- 主要言語
- Python
- スター
- 24.3k
- フォーク
- 4k
- 平均マージ
- 1日 1時間
- マージ済み PR(30日)
- 31
説明
Initial Checks
- I confirm that I'm using the latest version of MCP Python SDK
- I confirm that I searched for my issue in https://github.com/modelcontextprotocol/python-sdk/issues before opening this issue
Description
In the .well-known/oauth-authorization-server endpoint and , the issuer is forced to always contain a trailing slash e.g.,
https://your-mcp.com/instead ofhttps://your-mcp.com
as a byproduct of using pydantic'sAnyHttpUrltype.
This causes issues in both Google's ADK and IBM's MCP Context Forge because:
- when building the .well-known URL, they expect a discovery issuer URL that does not contain a trailing slash; and
- then they MUST verify that the returned metadata issuer URL is identical to the discovery issuer URL ("authorization server's issuer identifier value" in the spec) according to RFC 8414 Section 3.2; so
- when
OAuthMetadata.issuercontains the trailing slash, the discovery process is aborted.
OAuth 2.0 Authorization Server Metadata spec says that the client MUST remove trailing paths from when the issuer contains a path component:
If the issuer identifier value contains a path component, any
terminating "/" MUST be removed before inserting "/.well-known/" and
the well-known URI suffix between the host component and the path
component.
-- https://datatracker.ietf.org/doc/html/rfc8414#section-3.1
if the trailing / in https://example.com/ is a "path component", and should thus be stripped by the client, so I think the spec is ambiguous about the responsibilities of the client in the case where there the issuer identifier value contains a lone trailing slash.
I did note that the examples of issuer identifiers in the spec do not contain a lone trailing slash, i.e. they are https://example.com rather than https://example.com/.
For these reasons, and
- while it's listed as the client's responsibility to remove trailing slashes from the issuer identifier,
- I don't believe it's the server implementation's responsibility to intentionally make it harder for clients by returning a URL that do not follow the assumptions in the spec.
I think it's worth it to consider interpreting the spec as "the issuer field should not contain a trailing slash".
I also believe this issue could be similar in mechanism, but different in scope, to what is described in https://github.com/modelcontextprotocol/python-sdk/issues/1265
Example Code
# A demonstration on how AnyHttpUrl adds a trailing slash.
>>> from pydantic.networks import AnyHttpUrl
>>> x = AnyHttpUrl("http://localhost:8000")
>>> x
AnyHttpUrl('http://localhost:8000/')
>>> str(x)
'http://localhost:8000/'
>>>
Python & MCP Python SDK
Python 3.14
mcp==1.25.0
コントリビューションガイド
はじめの一歩
- issue を最後まで読み、次にプロジェクトのコントリビューションガイドを読みます。
- 着手することを issue にコメントします — 二人が同じ作業をするのを防げます。
- リポジトリをフォークし、ブランチを切って変更します。
- issue 番号を参照したプルリクエストを送ります。
調査の方向性
OAuthMetadataモデルとエンドポイント .well-known/oauth-authorization-server から始め、次にPydanticのAnyHttpUrlがissuer値をどのように正規化するかを調べてください。単独の末尾スラッシュについて期待される扱いを決める前に、RFC 8414の要件と既存のissueの議論を確認してください。返されるissuerとdiscovery URLが影響を受けるクライアントとの互換性を維持すれば完了です。
索引モデルが issue の本文から書いたものです。
評価
- 技術スタック
- python
- 領域
- api, authentication
- issue の種類
- バグ
- 難易度
- 3/5
- 見積もり時間
- 1〜2日
- 活発さ
- 活発
- 明瞭さ
- おおむね明確
- 初心者へのやさしさ
- 58/100