From c8866c59c8762783f96afc3bfa04791074cc9533 Mon Sep 17 00:00:00 2001
From: seanb4t <4678+seanb4t@users.noreply.github.com>
Date: Wed, 3 Jun 2026 10:05:51 -0400
Subject: [PATCH] fix(mcp): relay upstream OAuth errors from the token endpoint
 instead of masking as 500
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Fixes #29588.

The async HTTP handler raises on an upstream 4xx (MaskedHTTPStatusError) and
exchange_token_with_server doesn't catch it, so an upstream 400 invalid_grant
for an expired or rotated refresh token reaches the client as an opaque 500.
The client can't tell its refresh token is dead, so it never starts a new
authorization flow and retries the same token; headless clients loop.

Relay the upstream status and OAuth error body per RFC 6749 §5.2 via a new
relay_oauth_token_error helper in oauth_utils, matching the BYOK endpoint's
_oauth_token_error behavior. MaskedHTTPStatusError subclasses
httpx.HTTPStatusError and preserves the upstream status and body, so relaying
is exact.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
---
 .../mcp_server/discoverable_endpoints.py      | 18 +++--
 .../_experimental/mcp_server/oauth_utils.py   | 19 +++++
 .../mcp_server/test_discoverable_endpoints.py | 70 +++++++++++++++++++
 3 files changed, 102 insertions(+), 5 deletions(-)

diff --git a/litellm/proxy/_experimental/mcp_server/discoverable_endpoints.py b/litellm/proxy/_experimental/mcp_server/discoverable_endpoints.py
index ed374635fea..a13b040c5e0 100644
--- a/litellm/proxy/_experimental/mcp_server/discoverable_endpoints.py
+++ b/litellm/proxy/_experimental/mcp_server/discoverable_endpoints.py
@@ -17,6 +17,7 @@ from litellm.llms.custom_httpx.http_handler import (
 from litellm.proxy._experimental.mcp_server.oauth_utils import (
     TOKEN_NO_CACHE_HEADERS,
     get_request_base_url,
+    relay_oauth_token_error,
     validate_trusted_redirect_uri,
 )
 from litellm.proxy.auth.ip_address_utils import IPAddressUtils
@@ -456,11 +457,18 @@ async def exchange_token_with_server(
             token_data["code_verifier"] = code_verifier
 
     async_client = get_async_httpx_client(llm_provider=httpxSpecialProvider.Oauth2Check)
-    response = await async_client.post(
-        mcp_server.token_url,
-        headers={"Accept": "application/json"},
-        data=token_data,
-    )
+    try:
+        response = await async_client.post(
+            mcp_server.token_url,
+            headers={"Accept": "application/json"},
+            data=token_data,
+        )
+    except httpx.HTTPStatusError as e:
+        # The async HTTP handler raises on an upstream 4xx/5xx. Relay the OAuth
+        # error to the client (RFC 6749 §5.2) instead of letting it surface as a
+        # 500; otherwise the client can't see e.g. invalid_grant and start a new
+        # authorization flow, and a headless client loops on a dead refresh token.
+        return relay_oauth_token_error(e.response)
     if response is None:
         raise HTTPException(
             status_code=502,
diff --git a/litellm/proxy/_experimental/mcp_server/oauth_utils.py b/litellm/proxy/_experimental/mcp_server/oauth_utils.py
index e8b591c39cf..0c24e647a55 100644
--- a/litellm/proxy/_experimental/mcp_server/oauth_utils.py
+++ b/litellm/proxy/_experimental/mcp_server/oauth_utils.py
@@ -6,7 +6,9 @@ from ipaddress import ip_address
 from typing import Any, Dict, List, NoReturn, Optional
 from urllib.parse import ParseResult, urlparse, urlunparse
 
+import httpx
 from fastapi import HTTPException, Request
+from fastapi.responses import JSONResponse
 
 from litellm._logging import verbose_logger
 from litellm.proxy.auth.ip_address_utils import IPAddressUtils
@@ -65,6 +67,23 @@ def _oauth_invalid_request(
     raise HTTPException(status_code=400, detail=detail)
 
 
+def relay_oauth_token_error(response: httpx.Response) -> JSONResponse:
+    """Relay an upstream OAuth token-endpoint error to the client unchanged
+    (RFC 6749 §5.2). Masking it as a 500 hides the ``error`` field (e.g.
+    ``invalid_grant``) that clients rely on to drop a dead token and start a
+    new authorization flow.
+    """
+    try:
+        content = response.json()
+    except Exception:
+        content = {"error": "invalid_request", "error_description": response.text}
+    return JSONResponse(
+        status_code=response.status_code,
+        content=content,
+        headers=TOKEN_NO_CACHE_HEADERS,
+    )
+
+
 def _origin_label(scheme: str, netloc: str) -> str:
     """Human-readable origin for error messages (scheme + host[:port])."""
     return f"{scheme}://{netloc}" if netloc else f"{scheme}://"
diff --git a/tests/test_litellm/proxy/_experimental/mcp_server/test_discoverable_endpoints.py b/tests/test_litellm/proxy/_experimental/mcp_server/test_discoverable_endpoints.py
index da66d60aed8..a9cc2435cee 100644
--- a/tests/test_litellm/proxy/_experimental/mcp_server/test_discoverable_endpoints.py
+++ b/tests/test_litellm/proxy/_experimental/mcp_server/test_discoverable_endpoints.py
@@ -2661,3 +2661,73 @@ async def test_token_endpoint_sets_no_store_cache_control():
 
     assert response.headers["cache-control"] == "no-store"
     assert response.headers["pragma"] == "no-cache"
+
+
+@pytest.mark.asyncio
+async def test_exchange_token_relays_upstream_oauth_error():
+    """An upstream OAuth error (e.g. 400 invalid_grant for an expired refresh
+    token) is relayed to the client per RFC 6749 §5.2, not masked as a 500, so
+    the client can drop the dead token and start a new authorization flow."""
+    import json
+
+    import httpx
+    from fastapi import Request
+
+    from litellm.proxy._experimental.mcp_server.discoverable_endpoints import (
+        exchange_token_with_server,
+    )
+    from litellm.proxy._types import MCPTransport
+    from litellm.types.mcp import MCPAuth
+    from litellm.types.mcp_server.mcp_server_manager import MCPServer
+
+    server = MCPServer(
+        server_id="t",
+        name="t",
+        server_name="t",
+        alias="t",
+        transport=MCPTransport.http,
+        auth_type=MCPAuth.oauth2,
+        client_id="cid",
+        client_secret="cs",
+        authorization_url="https://provider.com/oauth/authorize",
+        token_url="https://provider.com/oauth/token",
+    )
+    mock_request = MagicMock(spec=Request)
+    mock_request.base_url = "https://litellm.example.com/"
+    mock_request.headers = {}
+
+    # The async HTTP handler raises on an upstream 4xx (it calls
+    # raise_for_status internally; MaskedHTTPStatusError subclasses
+    # httpx.HTTPStatusError and preserves the upstream status + body).
+    upstream = httpx.Response(
+        status_code=400,
+        json={"error": "invalid_grant"},
+        request=httpx.Request("POST", server.token_url),
+    )
+    fake_http_client = MagicMock()
+    fake_http_client.post = AsyncMock(
+        side_effect=httpx.HTTPStatusError(
+            "400 Bad Request", request=upstream.request, response=upstream
+        )
+    )
+
+    with patch(
+        "litellm.proxy._experimental.mcp_server.discoverable_endpoints.get_async_httpx_client",
+        return_value=fake_http_client,
+    ):
+        response = await exchange_token_with_server(
+            request=mock_request,
+            mcp_server=server,
+            grant_type="refresh_token",
+            code=None,
+            redirect_uri=None,
+            client_id="cid",
+            client_secret=None,
+            code_verifier=None,
+            refresh_token="expired-refresh-token",
+        )
+
+    assert response.status_code == 400
+    assert json.loads(response.body) == {"error": "invalid_grant"}
+    # Error responses must not be cached either.
+    assert response.headers["cache-control"] == "no-store"