diff --git a/litellm/proxy/_experimental/mcp_server/db.py b/litellm/proxy/_experimental/mcp_server/db.py index 082a90fdcfb..7379126983a 100644 --- a/litellm/proxy/_experimental/mcp_server/db.py +++ b/litellm/proxy/_experimental/mcp_server/db.py @@ -4,7 +4,7 @@ import hashlib import json from collections.abc import Awaitable, Callable, Iterable, Mapping, Sequence from datetime import datetime, timedelta, timezone -from typing import TYPE_CHECKING, Any, Final, Protocol, TypedDict, cast +from typing import TYPE_CHECKING, Any, Final, Literal, Protocol, TypedDict, cast from litellm._logging import verbose_proxy_logger from litellm._uuid import uuid @@ -125,6 +125,9 @@ class OAuthCredentialPayload(_OAuthCredentialAccessToken, total=False): server_id: str +OAuthGrantState = Literal["valid", "refreshable", "absent"] + + class _OAuthTokenRefreshResponse(TypedDict, total=False): access_token: str refresh_token: str @@ -1465,6 +1468,15 @@ def is_oauth_credential_expired(cred: OAuthCredentialPayload, buffer_seconds: in return False +def oauth_grant_state(cred: OAuthCredentialPayload | None) -> OAuthGrantState: + """Classify local grant readiness without attempting a refresh or checking upstream revocation.""" + if not cred or not cred.get("access_token"): + return "absent" + if not is_oauth_credential_expired(cred, buffer_seconds=MCP_PER_USER_TOKEN_EXPIRY_BUFFER_SECONDS): + return "valid" + return "refreshable" if cred.get("refresh_token") else "absent" + + async def get_user_oauth_credential( prisma_client: PrismaClient, user_id: str, @@ -1727,12 +1739,11 @@ async def resolve_valid_user_oauth_token( dict it already holds. ``prisma_client`` is fetched lazily and only when a refresh actually happens, so the valid-token path never requires a DB handle. """ - if not cred or not cred.get("access_token"): + grant: Final = oauth_grant_state(cred) + if cred is None or grant == "absent": return None - if not is_oauth_credential_expired(cred, buffer_seconds=MCP_PER_USER_TOKEN_EXPIRY_BUFFER_SECONDS): + if grant == "valid": return cred - if not cred.get("refresh_token"): - return None if prisma_client is None: from litellm.proxy.utils import get_prisma_client_or_throw diff --git a/litellm/proxy/_experimental/mcp_server/discoverable_endpoints.py b/litellm/proxy/_experimental/mcp_server/discoverable_endpoints.py index e10bfd41ed6..cab4b6c161a 100644 --- a/litellm/proxy/_experimental/mcp_server/discoverable_endpoints.py +++ b/litellm/proxy/_experimental/mcp_server/discoverable_endpoints.py @@ -43,9 +43,11 @@ from litellm.proxy._experimental.mcp_server.faults import ( render_token_fault, ) from litellm.proxy._experimental.mcp_server.gateway_dcr_flow import ( + VendorCredentialState, aggregate_authorize, aggregate_token, complete_connect_flow, + describe_connect_flow, introspect_gateway_token, is_gateway_dcr_client_id, is_proxy_api_resource, @@ -798,21 +800,7 @@ def _bridge_access_denied_redirect(redirect_uri: str, state: str, mcp_server: MC return RedirectResponse(_append_query_params(redirect_uri, params), status_code=302) -async def _bridge_authorize_access_denial( - litellm_user_id: str, - mcp_server: MCPServer, - redirect_uri: str, - state: str, -) -> RedirectResponse | None: - """The denial redirect for a signed-in user who cannot reach the target server, or None to proceed. - - Admits the user exactly as MCP egress will (the same ``reload_admitted_user`` constructor and the - same ``get_allowed_mcp_servers`` resolver), so an envelope is minted only when the resulting - session can actually list and call the server's tools. Without this gate the flow completes, the - client shows connected, and every tool request fail-closes to an empty list with nothing telling - the operator why. An availability fault (5xx, e.g. a DB outage's 503) propagates; an unknown or - deactivated user denies like a missing grant, fail closed. - """ +async def _user_can_reach_mcp_server(user_id: str, server_id: str) -> bool: from litellm.proxy._experimental.mcp_server.auth.user_api_key_auth_mcp import ( MCPRequestHandler, ) @@ -821,13 +809,22 @@ async def _bridge_authorize_access_denial( ) try: - admitted: Final = await MCPRequestHandler.reload_admitted_user(litellm_user_id) + admitted: Final = await MCPRequestHandler.reload_admitted_user(user_id) except HTTPException as exc: if exc.status_code >= 500: raise - return _bridge_access_denied_redirect(redirect_uri, state, mcp_server) - allowed_server_ids: Final = await global_mcp_server_manager.get_allowed_mcp_servers(admitted) - if mcp_server.server_id in allowed_server_ids: + return False + return server_id in await global_mcp_server_manager.get_allowed_mcp_servers(admitted) + + +async def _bridge_authorize_access_denial( + litellm_user_id: str, + mcp_server: MCPServer, + redirect_uri: str, + state: str, +) -> RedirectResponse | None: + """The denial redirect for a signed-in user who cannot reach the target server, or None to proceed.""" + if await _user_can_reach_mcp_server(litellm_user_id, mcp_server.server_id): return None return _bridge_access_denied_redirect(redirect_uri, state, mcp_server) @@ -1910,6 +1907,38 @@ async def token_endpoint( ) +async def _vendor_credential_state(user_id: str, server_id: str) -> VendorCredentialState: + """Whether the gateway itself can see a live vendor credential for this user and server. + + The one reading of "authorized" the connect page displays and the finish step enforces, so + the button a user sees and the grant they get cannot disagree. A read fault is neither, and + fails the scoped grant closed.""" + from litellm.proxy._experimental.mcp_server.db import ( # noqa: PLC0415 # circular import at module load + get_user_oauth_credential, + oauth_grant_state, + ) + from litellm.proxy.proxy_server import prisma_client # noqa: PLC0415 # circular import at module load + + if prisma_client is None: + return "unavailable" + try: + credential: Final = await get_user_oauth_credential(prisma_client, user_id, server_id) + except Exception: # noqa: BLE001 # a credential-read fault must fail the scoped grant closed + return "unavailable" + return "absent" if oauth_grant_state(credential) == "absent" else "present" + + +@router.get("/authorize/flow") +async def authorize_flow(request: Request, flow: str) -> Response: + return await describe_connect_flow( + request=request, + flow_handle=flow, + session_user_id=_session_cookie_user_id(request), + lookup_vendor_credential=_vendor_credential_state, + lookup_server_reachability=_user_can_reach_mcp_server, + ) + + @router.post("/authorize/complete") async def authorize_complete( request: Request, @@ -1934,6 +1963,8 @@ async def authorize_complete( delivery=delivery, team_id=team_id, decision=decision, + lookup_vendor_credential=_vendor_credential_state, + lookup_server_reachability=_user_can_reach_mcp_server, ) diff --git a/litellm/proxy/_experimental/mcp_server/gateway_dcr_flow.py b/litellm/proxy/_experimental/mcp_server/gateway_dcr_flow.py index c7b0045dde5..3d94fa345d0 100644 --- a/litellm/proxy/_experimental/mcp_server/gateway_dcr_flow.py +++ b/litellm/proxy/_experimental/mcp_server/gateway_dcr_flow.py @@ -152,10 +152,8 @@ _AUTH_CODE_DEBUG_KEY: Final = "gateway_authorization_code" ReloadUserFailure = Literal["unresolvable", "unavailable", "faulted", "no_active_key"] ReloadUser = Callable[[str], Awaitable[ReloadUserFailure | None]] -"""Injected live-user revalidation (the token endpoint's mirror of admission): -``None`` means the user is active; ``unavailable`` is a retryable DB outage; ``faulted`` is -a DB fault retrying will not clear (still 503, worded so nobody just waits); anything else -fails the grant closed.""" +VendorCredentialState = Literal["present", "absent", "unavailable"] +"""The per-user vendor credential read has three outcomes: present, absent, or unavailable.""" _DB_UNAVAILABLE_DESCRIPTION: Final = "the gateway database is unavailable; retry" _DB_FAULTED_DESCRIPTION: Final = ( @@ -195,6 +193,16 @@ class ConsentTeam(BaseModel): team_alias: str | None = None +class LookupVendorCredential(Protocol): + """Injected read of a user's vendor credential for one server.""" + + def __call__(self, user_id: str, server_id: str, /) -> Awaitable[VendorCredentialState]: ... + + +class LookupServerReachability(Protocol): + def __call__(self, user_id: str, server_id: str, /) -> Awaitable[bool]: ... + + class LookupConsentTeams(Protocol): """Injected lookup of the teams a signed-in user may bind a proxy-API credential to.""" @@ -205,6 +213,14 @@ async def _refuse_proxy_credential(user_id: str, team_id: str | None) -> ProxyCr return "unresolvable" +async def _unavailable_vendor_credential(user_id: str, server_id: str) -> VendorCredentialState: + return "unavailable" + + +async def _unreachable_server(user_id: str, server_id: str) -> bool: + return False + + class GatewayDcrClient(BaseModel): """The registration record sealed into a gateway DCR ``client_id``. @@ -449,7 +465,10 @@ def aggregate_authorize( A per-server RFC 8707 ``resource`` naming a gateway-managed oauth2 server scopes the flow to that one server: the scope is sealed into the flow, carried into the code, and - bound into the session token, while the connect page interlude runs exactly as before. + bound into the session token. The connect URL carries only the flow handle; the page + learns the client origin, the scoped server, and whether its vendor OAuth is done from + :func:`describe_connect_flow`, which reads the sealed flow, so nothing a link can carry + steers which server the page authorizes or names on the confirmation. Validation failures respond directly with 400 and never redirect: per RFC 6749 section 4.1.2.1 an unvalidated redirect URI must not receive an error redirect, and @@ -474,10 +493,7 @@ def aggregate_authorize( resource_server_id=scoped_server.server_id if scoped_server is not None else None, audience=None, ) - connect_url: Final = _append_query_params( - f"{base_url}/ui/connect", - (("connect_flow", handle), ("connect_client", _origin_only(redirect_uri))), - ) + connect_url: Final = _append_query_params(f"{base_url}/ui/connect", (("connect_flow", handle),)) response: Final = RedirectResponse(connect_url, status_code=303) _set_flow_cookie(response, request, handle, flow) return response @@ -684,6 +700,99 @@ def _origin_only(url: str) -> str: return f"{parsed.scheme}://{parsed.netloc}" if parsed.netloc else "" +def _open_flow_for( + request: Request, flow_handle: str, session_user_id: str | None, now: datetime +) -> _ConnectFlow | Response: + sealed_flow: Final = request.cookies.get(_flow_cookie_name(flow_handle)) + if sealed_flow is None: + return _oauth_error(400, "invalid_request", "unknown or expired connect flow") + flow: Final = _open_sealed(sealed_flow, _UNPREFIXED, _ConnectFlow, _CONNECT_FLOW_DEBUG_KEY) + if flow is None or now.timestamp() >= flow.exp: + return _oauth_error(400, "invalid_request", "unknown or expired connect flow") + if session_user_id is None: + return _oauth_error(401, "login_required", "sign in to LiteLLM to finish connecting") + if session_user_id != flow.user_id: + return _oauth_error(403, "access_denied", "the signed-in user does not match this connect flow") + return flow + + +async def _flow_target( + flow: _ConnectFlow, lookup_server_reachability: LookupServerReachability +) -> tuple[Literal["unscoped", "interactive", "m2m", "stale"], MCPServer | None]: + if flow.resource_server_id is None: + return "unscoped", None + from litellm.proxy._experimental.mcp_server.mcp_server_manager import ( # noqa: PLC0415 # import cycle + MCPServerManager, + global_mcp_server_manager, + ) + + server: Final = global_mcp_server_manager.get_mcp_server_by_id(flow.resource_server_id) + if ( + server is None + or not server.is_gateway_managed_oauth2 + or not await lookup_server_reachability(flow.user_id, server.server_id) + ): + return "stale", None + state: Final = "m2m" if MCPServerManager.effective_oauth2_flow(server) == "client_credentials" else "interactive" + return state, server + + +class ConnectFlowDescription(TypedDict): + """What the connect page is allowed to know about one in-flight flow.""" + + state: ReadOnly[Literal["unscoped", "interactive", "m2m", "stale"]] + client_origin: ReadOnly[str] + server_id: ReadOnly[str | None] + server_name: ReadOnly[str | None] + connected: ReadOnly[bool | None] + + +async def _describe_opened_flow( + flow: _ConnectFlow, + lookup_vendor_credential: LookupVendorCredential, + lookup_server_reachability: LookupServerReachability, +) -> ConnectFlowDescription | Response: + state, server = await _flow_target(flow, lookup_server_reachability) + if state == "interactive" and server is not None: + credential: Final = await lookup_vendor_credential(flow.user_id, server.server_id) + if credential == "unavailable": + return _oauth_error(503, "temporarily_unavailable", _DB_UNAVAILABLE_DESCRIPTION) + interactive_description: Final[ConnectFlowDescription] = { + "state": state, + "client_origin": _origin_only(flow.redirect_uri), + "server_id": server.server_id, + "server_name": server.server_name or server.alias or server.name, + "connected": credential == "present", + } + return interactive_description + described: Final[ConnectFlowDescription] = { + "state": state, + "client_origin": _origin_only(flow.redirect_uri), + "server_id": None if server is None else server.server_id, + "server_name": None if server is None else (server.server_name or server.alias or server.name), + "connected": state == "m2m" or None, + } + return described + + +async def describe_connect_flow( + request: Request, + flow_handle: str, + session_user_id: str | None, + lookup_vendor_credential: LookupVendorCredential, + lookup_server_reachability: LookupServerReachability, +) -> Response: + opened: Final = _open_flow_for(request, flow_handle, session_user_id, datetime.now(timezone.utc)) + if isinstance(opened, Response): + return opened + described: Final = await _describe_opened_flow(opened, lookup_vendor_credential, lookup_server_reachability) + return ( + described + if isinstance(described, Response) + else JSONResponse(content=described, headers=TOKEN_NO_CACHE_HEADERS) + ) + + async def complete_connect_flow( request: Request, flow_handle: str, @@ -692,56 +801,34 @@ async def complete_connect_flow( delivery: str | None = None, team_id: str | None = None, decision: str | None = None, + lookup_vendor_credential: LookupVendorCredential = _unavailable_vendor_credential, + lookup_server_reachability: LookupServerReachability = _unreachable_server, ) -> Response: - """The deliberate finish step of the connect flow: mint the gateway authorization - code and send the browser back to the client. + """Mint the code only after a deliberate POST by the sealed user. - Reached by POST so a cross-site GET cannot trigger it, and bound to the HttpOnly - per-flow cookie plus an exact match between the signed-in user and the user sealed - into the flow: a link crafted by another party dies here with ``access_denied`` - instead of minting a code for the victim's identity. The flow is single-use (an atomic - claim on its ``jti``), so a double-submit cannot mint two codes from one sign-in. - - ``delivery`` chooses how the code reaches the client. Default (absent or - ``"redirect"``) is the 303 to the client's registered redirect URI. ``"manual"`` - renders the callback URL on a page instead, for a client whose redirect URI is a - loopback host but which runs on a DIFFERENT machine than the browser (EC2/SSH box, - container): the 303 would dereference the browser machine's loopback and the code - would never arrive, so the user carries it over by pasting the URL into the client or - fetching it from the client machine's terminal. Manual delivery is honored only for - loopback redirect URIs; a routable redirect URI works from any browser by - construction, so those flows always redirect. The user who sees the page is exactly - the user the 303 would have carried the code to, and the same user already sees the - code today in the dead redirect's address bar, so the page exposes the code to no new - party. Unknown ``delivery`` values are rejected rather than defaulted: a client that - asked for manual delivery and got a dead redirect instead would silently lose its - code. - - ``decision`` and ``team_id`` come from the native-client consent page. ``"deny"`` - burns the flow and sends the client ``error=access_denied`` so it stops waiting; - ``team_id`` is sealed into the code only for proxy-API flows, where it picks which of - the user's teams the minted credential is attributed to. + A scoped flow additionally requires its sealed server to have a live vendor credential + before a code can be minted. The check happens before the single-use claim, so a + premature submit can be retried after authorization; denial deliberately bypasses it. """ if delivery not in (None, "redirect", "manual"): return _oauth_error(400, "invalid_request", "delivery must be 'redirect' or 'manual'") if decision not in (None, "approve", "deny"): return _oauth_error(400, "invalid_request", "decision must be 'approve' or 'deny'") - sealed_flow: Final = request.cookies.get(_flow_cookie_name(flow_handle)) - if sealed_flow is None: - return _oauth_error(400, "invalid_request", "unknown or expired connect flow") - flow: Final = _open_sealed(sealed_flow, _UNPREFIXED, _ConnectFlow, _CONNECT_FLOW_DEBUG_KEY) - if flow is None: - return _oauth_error(400, "invalid_request", "unknown or expired connect flow") now: Final = datetime.now(timezone.utc) - if now.timestamp() >= flow.exp: - return _oauth_error(400, "invalid_request", "the connect flow has expired; restart the connection") - if session_user_id is None: - return _oauth_error(401, "login_required", "sign in to LiteLLM to finish connecting") - if session_user_id != flow.user_id: - return _oauth_error(403, "access_denied", "the signed-in user does not match this connect flow") + opened: Final = _open_flow_for(request, flow_handle, session_user_id, now) + if isinstance(opened, Response): + return opened + if decision != "deny": + described: Final = await _describe_opened_flow(opened, lookup_vendor_credential, lookup_server_reachability) + if isinstance(described, Response): + return described + if described["state"] == "stale": + return _oauth_error(400, "invalid_request", "the requested MCP server is no longer available") + if described["connected"] is False: + return _oauth_error(400, "invalid_request", "authorize the requested MCP server before finishing") flow_refusal: Final = _claim_refusal( await _SingleUseGuard(cache).claim( - f"{_USED_FLOW_CACHE_PREFIX}{flow.jti}", CONNECT_FLOW_TTL_SECONDS + _CLAIM_TTL_BUFFER_SECONDS + f"{_USED_FLOW_CACHE_PREFIX}{opened.jti}", CONNECT_FLOW_TTL_SECONDS + _CLAIM_TTL_BUFFER_SECONDS ), replayed=_oauth_error( 400, "invalid_request", "this connect flow was already completed; restart the connection" @@ -750,7 +837,7 @@ async def complete_connect_flow( if flow_refusal is not None: return flow_refusal response: Final = ( - _denied_flow_response(flow) if decision == "deny" else _approved_flow_response(flow, delivery, team_id, now) + _denied_flow_response(opened) if decision == "deny" else _approved_flow_response(opened, delivery, team_id, now) ) path, secure = _cookie_path_and_secure(request) response.delete_cookie(key=_flow_cookie_name(flow_handle), path=path, secure=secure, httponly=True, samesite="lax") diff --git a/litellm/proxy/_lazy_openapi_snapshot.json b/litellm/proxy/_lazy_openapi_snapshot.json index dba2b2428aa..71475320c2c 100644 --- a/litellm/proxy/_lazy_openapi_snapshot.json +++ b/litellm/proxy/_lazy_openapi_snapshot.json @@ -19886,6 +19886,46 @@ ] } }, + "/authorize/flow": { + "get": { + "operationId": "authorize_flow_authorize_flow_get", + "parameters": [ + { + "in": "query", + "name": "flow", + "required": true, + "schema": { + "title": "Flow", + "type": "string" + } + } + ], + "responses": { + "200": { + "content": { + "application/json": { + "schema": {} + } + }, + "description": "Successful Response" + }, + "422": { + "content": { + "application/json": { + "schema": { + "$ref": "#/components/schemas/HTTPValidationError" + } + } + }, + "description": "Validation Error" + } + }, + "summary": "Authorize Flow", + "tags": [ + "mcp_discoverable" + ] + } + }, "/callback": { "get": { "description": "OAuth 2.0 authorization response handler for MCP loopback clients.\n\nAccepts either:\n\n- A successful authorization response (``code`` + ``state``), which is\n forwarded back to the validated client ``redirect_uri`` with the\n original (un-wrapped) ``state``.\n- An error response (``error``[+``error_description``/``error_uri``]), per\n RFC 6749 \u00a74.1.2.1. When ``state`` is present and decodes to a trusted\n ``redirect_uri``, the error params are propagated back to the client so\n its OAuth library can surface them. Otherwise we render an HTML error\n page so the user is not left on an opaque 422 / blank screen.", 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 be4206a1faf..763200c3709 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 @@ -4,6 +4,7 @@ import hashlib import json import time from base64 import urlsafe_b64encode +from datetime import datetime, timedelta, timezone from typing import TYPE_CHECKING from unittest.mock import AsyncMock, MagicMock, patch @@ -18,6 +19,57 @@ if TYPE_CHECKING: from litellm.types.mcp_server.mcp_server_manager import MCPServer +def _stored_grant(access_token="access-token", refresh_token=None, expires_in_seconds=None, expires_at=None): + credential = {"type": "oauth2", "access_token": access_token} + if refresh_token is not None: + credential["refresh_token"] = refresh_token + if expires_in_seconds is not None: + credential["expires_at"] = (datetime.now(timezone.utc) + timedelta(seconds=expires_in_seconds)).isoformat() + if expires_at is not None: + credential["expires_at"] = expires_at + return credential + + +@pytest.mark.asyncio +@pytest.mark.parametrize( + ("fields", "egress_has_token"), + [ + (None, False), + ({"access_token": "", "refresh_token": "refresh-token"}, False), + ({}, True), + ({"expires_at": "never"}, True), + ({"expires_in_seconds": 600}, True), + ({"expires_in_seconds": 30}, False), + ({"expires_in_seconds": -300}, False), + ({"expires_in_seconds": -300, "refresh_token": ""}, False), + ({"expires_in_seconds": 30, "refresh_token": "refresh-token"}, True), + ({"expires_in_seconds": -300, "refresh_token": "refresh-token"}, True), + ], +) +async def test_vendor_credential_state_agrees_with_egress_token_resolution(monkeypatch, fields, egress_has_token): + from litellm.proxy._experimental.mcp_server import db as mcp_db + from litellm.proxy._experimental.mcp_server import discoverable_endpoints + + monkeypatch.setattr(mcp_db, "MCP_PER_USER_TOKEN_EXPIRY_BUFFER_SECONDS", 60) + credential = _stored_grant(**fields) if fields is not None else None + read = AsyncMock(return_value=credential) + refresh = AsyncMock(return_value=_stored_grant(access_token="fresh-token", expires_in_seconds=3600)) + prisma = MagicMock() + monkeypatch.setattr("litellm.proxy.proxy_server.prisma_client", prisma) + monkeypatch.setattr(mcp_db, "get_user_oauth_credential", read) + monkeypatch.setattr(mcp_db, "refresh_user_oauth_token", refresh) + + connect = await discoverable_endpoints._vendor_credential_state("user-1", "server-1") + read.assert_awaited_once_with(prisma, "user-1", "server-1") + refresh.assert_not_awaited() + egress = await mcp_db.resolve_valid_user_oauth_token( + user_id="user-1", server=MagicMock(), cred=credential, prisma_client=prisma + ) + + assert (egress is not None) is egress_has_token + assert connect == ("present" if egress_has_token else "absent") + + # Fixture to mock IP address check for all MCP tests # This prevents tests from failing due to IP-based access control @pytest.fixture(autouse=True) diff --git a/tests/test_litellm/proxy/_experimental/mcp_server/test_gateway_dcr_flow.py b/tests/test_litellm/proxy/_experimental/mcp_server/test_gateway_dcr_flow.py index 1670370f082..73a52a8d2e8 100644 --- a/tests/test_litellm/proxy/_experimental/mcp_server/test_gateway_dcr_flow.py +++ b/tests/test_litellm/proxy/_experimental/mcp_server/test_gateway_dcr_flow.py @@ -26,6 +26,7 @@ from litellm.proxy._experimental.mcp_server.gateway_dcr_flow import ( aggregate_authorize, aggregate_token, complete_connect_flow, + describe_connect_flow, introspect_gateway_token, is_gateway_dcr_client_id, is_proxy_api_resource, @@ -230,7 +231,7 @@ async def test_authorize_with_session_hands_browser_to_connect_page_with_flow_co assert location.path == "/ui/connect" params = parse_qs(location.query) handle = params["connect_flow"][0] - assert params["connect_client"] == ["https://claude.ai"] + assert set(params) == {"connect_flow"} set_cookie = response.headers["set-cookie"] assert f"{CONNECT_FLOW_COOKIE_PREFIX}{handle}" in set_cookie assert "HttpOnly" in set_cookie @@ -889,14 +890,60 @@ def _opened_principal(payload): return admitted.principal -async def _finish_connect_page(response): +class _VendorCredential: + def __init__(self, state="present"): + self.calls = [] + self.state = state + + async def __call__(self, user_id, server_id): + self.calls.append((user_id, server_id)) + return self.state + + +class _ServerReachability: + def __init__(self, reachable=True): + self.calls = [] + self.reachable = reachable + + async def __call__(self, user_id, server_id): + self.calls.append((user_id, server_id)) + return self.reachable + + +async def _complete_page(response, scoped_server=None, vendor=None, reachable=None, cache=None, **overrides): + from unittest.mock import patch + handle, cookies = _flow_cookie_from(response) - completed = await complete_connect_flow( - request=_request("/authorize/complete", cookies=cookies, method="POST"), - flow_handle=handle, - session_user_id="u1", - cache=DualCache(), - ) + with patch(_MANAGER_PATCH) as manager: + manager.get_mcp_server_by_id.return_value = scoped_server + return await complete_connect_flow( + request=_request("/authorize/complete", cookies=cookies, method="POST"), + flow_handle=handle, + session_user_id="u1", + cache=cache or DualCache(), + lookup_vendor_credential=vendor or _VendorCredential(), + lookup_server_reachability=reachable or _ServerReachability(), + **overrides, + ) + + +async def _describe_page(response, scoped_server=None, vendor=None, reachable=None, session_user_id="u1", cookies=None): + from unittest.mock import patch + + handle, flow_cookies = _flow_cookie_from(response) + with patch(_MANAGER_PATCH) as manager: + manager.get_mcp_server_by_id.return_value = scoped_server + return await describe_connect_flow( + request=_request("/authorize/flow", cookies=flow_cookies if cookies is None else cookies), + flow_handle=handle, + session_user_id=session_user_id, + lookup_vendor_credential=vendor or _VendorCredential(), + lookup_server_reachability=reachable or _ServerReachability(), + ) + + +async def _finish_connect_page(response, scoped_server=None): + completed = await _complete_page(response, scoped_server=scoped_server) return parse_qs(urlparse(completed.headers["location"]).query)["code"][0] @@ -910,23 +957,43 @@ def _sealed_wire_json(sealed, prefix, debug_key): @pytest.mark.asyncio async def test_scoped_authorize_runs_connect_page_with_sealed_scope(): - """LIT-4917: a per-server RFC 8707 resource naming a gateway-managed oauth2 server - seals that server into the flow. The connect page interlude runs exactly as before - (the scope restricts, it never skips consent), and the code minted at the finish step - and the session pair it redeems for are both scoped.""" + """LIT-4917 plus LIT-7075: a per-server RFC 8707 resource naming a gateway-managed oauth2 + server seals that server into the flow. The connect URL carries only the handle; the page + learns the scoped server and its vendor state from describe_connect_flow, and the finish + step refuses to mint a scoped code until that vendor credential exists, without burning + the flow. The code minted afterwards and the session pair it redeems for are both scoped.""" from unittest.mock import patch client_id = (await _register([REDIRECT_URI]))["client_id"] + github = _scoped_mcp_server() with patch(_MANAGER_PATCH) as manager: - manager.get_mcp_server_by_name.return_value = _scoped_mcp_server() + manager.get_mcp_server_by_name.return_value = github response = _scoped_authorize(client_id, SCOPED_RESOURCE) assert response.status_code == 303 - assert "/ui/connect" in response.headers["location"] + location = urlparse(response.headers["location"]) + assert location.path == "/ui/connect" + assert set(parse_qs(location.query)) == {"connect_flow"} _, cookies = _flow_cookie_from(response) assert ( _sealed_wire_json(next(iter(cookies.values())), "", "gateway_connect_flow")["resource_server_id"] == "github-id" ) - code = await _finish_connect_page(response) + described = await _describe_page(response, scoped_server=github, vendor=_VendorCredential("absent")) + assert json.loads(described.body) == { + "state": "interactive", + "client_origin": "https://claude.ai", + "server_id": "github-id", + "server_name": "github", + "connected": False, + } + cache = DualCache() + premature = await _complete_page(response, scoped_server=github, vendor=_VendorCredential("absent"), cache=cache) + assert premature.status_code == 400 + assert "authorize the requested MCP server" in json.loads(premature.body)["error_description"] + present = _VendorCredential("present") + completed = await _complete_page(response, scoped_server=github, vendor=present, cache=cache) + assert completed.status_code == 303 + assert present.calls == [("u1", "github-id")] + code = parse_qs(urlparse(completed.headers["location"]).query)["code"][0] assert ( _sealed_wire_json(code, GATEWAY_AUTH_CODE_PREFIX, "gateway_authorization_code")["resource_server_id"] == "github-id" @@ -952,9 +1019,11 @@ async def test_scoped_authorize_runs_connect_page_with_sealed_scope(): ) async def test_unscoped_resources_leave_flow_and_token_byte_identical(resource, resolves): """Every resource shape outside 'exactly one gateway-managed server' keeps today's flow: - connect page interlude, and NONE of the minted artifacts carry the scope key on the - wire, not the flow cookie, not the code, not the session JWT, so an unscoped flow - started on a new pod completes on a pod whose strict models predate the claim.""" + the generic connect grid (describe names no server, the finish step never consults the + vendor credential), and NONE of the minted + artifacts carry the scope key on the wire, not the flow cookie, not the code, not the + session JWT, so an unscoped flow started on a new pod completes on a pod whose strict + models predate the claim.""" import base64 from unittest.mock import patch @@ -963,10 +1032,18 @@ async def test_unscoped_resources_leave_flow_and_token_byte_identical(resource, manager.get_mcp_server_by_name.return_value = None if resolves is None else _scoped_mcp_server() response = _scoped_authorize(client_id, resource) assert response.status_code == 303 - assert "/ui/connect" in response.headers["location"] + location = urlparse(response.headers["location"]) + assert location.path == "/ui/connect" + assert set(parse_qs(location.query)) == {"connect_flow"} _, cookies = _flow_cookie_from(response) assert "resource_server_id" not in _sealed_wire_json(next(iter(cookies.values())), "", "gateway_connect_flow") - code = await _finish_connect_page(response) + vendor = _VendorCredential("absent") + described = await _describe_page(response, vendor=vendor) + assert json.loads(described.body)["state"] == "unscoped" + assert json.loads(described.body)["server_id"] is None + completed = await _complete_page(response, vendor=vendor) + assert vendor.calls == [] + code = parse_qs(urlparse(completed.headers["location"]).query)["code"][0] assert "resource_server_id" not in _sealed_wire_json(code, GATEWAY_AUTH_CODE_PREFIX, "gateway_authorization_code") token_response = await _redeem(code, client_id) payload = json.loads(token_response.body) @@ -979,19 +1056,150 @@ async def test_unscoped_resources_leave_flow_and_token_byte_identical(resource, @pytest.mark.asyncio async def test_scoped_authorize_delegate_server_stays_unscoped(): """A delegate-auth oauth2 server is outside the gateway-managed set (its keyless flow is - upstream PKCE via the relay), so a resource naming it never scopes the gateway flow.""" + upstream PKCE via the relay), so a resource naming it never scopes the gateway flow and + never narrows the connect page to it.""" from unittest.mock import patch client_id = (await _register([REDIRECT_URI]))["client_id"] with patch(_MANAGER_PATCH) as manager: manager.get_mcp_server_by_name.return_value = _scoped_mcp_server(delegate_auth_to_upstream=True) response = _scoped_authorize(client_id, SCOPED_RESOURCE) - assert "/ui/connect" in response.headers["location"] + location = urlparse(response.headers["location"]) + assert location.path == "/ui/connect" + assert set(parse_qs(location.query)) == {"connect_flow"} + assert json.loads((await _describe_page(response)).body)["server_id"] is None code = await _finish_connect_page(response) token_response = await _redeem(code, client_id) assert _opened_principal(json.loads(token_response.body)).resource_server_id is None +@pytest.mark.asyncio +async def test_m2m_scoped_flow_mints_without_a_user_credential(): + """A client-credentials server is already authorized by its gateway service credential, so + a resource-scoped flow finishes without consulting the per-user vault.""" + from unittest.mock import patch + + client_id = (await _register([REDIRECT_URI]))["client_id"] + m2m = _scoped_mcp_server(oauth2_flow="client_credentials") + with patch(_MANAGER_PATCH) as manager: + manager.get_mcp_server_by_name.return_value = m2m + response = _scoped_authorize(client_id, SCOPED_RESOURCE) + vendor = _VendorCredential("unavailable") + described = await _describe_page(response, scoped_server=m2m, vendor=vendor) + assert json.loads(described.body)["state"] == "m2m" + assert json.loads(described.body)["connected"] is True + assert vendor.calls == [] + completed = await _complete_page(response, scoped_server=m2m, vendor=vendor) + assert completed.status_code == 303 + assert vendor.calls == [] + + +@pytest.mark.asyncio +@pytest.mark.parametrize("oauth2_flow", ["authorization_code", "client_credentials"]) +async def test_unreachable_scoped_flow_cannot_describe_or_finish(oauth2_flow): + from unittest.mock import patch + + client_id = (await _register([REDIRECT_URI]))["client_id"] + server = _scoped_mcp_server(oauth2_flow=oauth2_flow) + with patch(_MANAGER_PATCH) as manager: + manager.get_mcp_server_by_name.return_value = server + response = _scoped_authorize(client_id, SCOPED_RESOURCE) + reachable = _ServerReachability(False) + vendor = _VendorCredential("present") + described = await _describe_page(response, scoped_server=server, reachable=reachable, vendor=vendor) + assert json.loads(described.body) == { + "state": "stale", + "client_origin": "https://claude.ai", + "server_id": None, + "server_name": None, + "connected": None, + } + cache = DualCache() + refused = await _complete_page(response, scoped_server=server, reachable=reachable, vendor=vendor, cache=cache) + assert refused.status_code == 400 + assert vendor.calls == [] + assert reachable.calls == [("u1", "github-id"), ("u1", "github-id")] + completed = await _complete_page(response, scoped_server=server, vendor=vendor, cache=cache) + assert completed.status_code == 303 + + +@pytest.mark.asyncio +async def test_stale_scoped_flow_remains_distinct_from_unscoped(): + """A server removed after authorize stays a stale scoped flow, so the page cannot offer a + broader unscoped grant or report a misleading Finish action.""" + from unittest.mock import patch + + client_id = (await _register([REDIRECT_URI]))["client_id"] + with patch(_MANAGER_PATCH) as manager: + manager.get_mcp_server_by_name.return_value = _scoped_mcp_server() + response = _scoped_authorize(client_id, SCOPED_RESOURCE) + described = await _describe_page(response, scoped_server=None) + assert json.loads(described.body)["state"] == "stale" + assert json.loads(described.body)["connected"] is None + stale = await _complete_page(response, scoped_server=None) + assert stale.status_code == 400 + assert json.loads(stale.body)["error_description"] == "the requested MCP server is no longer available" + + +@pytest.mark.asyncio +async def test_scoped_flow_deny_and_stale_server_never_need_the_vendor_credential(): + """Cancel is the escape hatch: a scoped user who cannot finish the vendor step still ends + the flow with access_denied and no credential lookup. A scoped server that is no longer + gateway-managed refuses to mint (nothing could serve that code) but also burns nothing.""" + from unittest.mock import patch + + client_id = (await _register([REDIRECT_URI]))["client_id"] + github = _scoped_mcp_server() + with patch(_MANAGER_PATCH) as manager: + manager.get_mcp_server_by_name.return_value = github + response = _scoped_authorize(client_id, SCOPED_RESOURCE) + cache = DualCache() + skipped_reachability = _ServerReachability(False) + stale = await _complete_page(response, scoped_server=None, reachable=skipped_reachability, cache=cache) + assert stale.status_code == 400 + assert skipped_reachability.calls == [] + assert json.loads(stale.body)["error_description"] == "the requested MCP server is no longer available" + described = await _describe_page(response, scoped_server=github, vendor=_VendorCredential("unavailable")) + assert described.status_code == 503 + vendor = _VendorCredential("absent") + deny_reachability = _ServerReachability(False) + denied = await _complete_page( + response, + scoped_server=github, + vendor=vendor, + reachable=deny_reachability, + cache=cache, + decision="deny", + ) + assert denied.status_code == 303 + assert parse_qs(urlparse(denied.headers["location"]).query)["error"] == ["access_denied"] + assert vendor.calls == [] + assert deny_reachability.calls == [] + + +@pytest.mark.asyncio +@pytest.mark.parametrize( + "session_user_id, cookies, expected_status, expected_error", + [ + ("u1", {}, 400, "invalid_request"), + ("u1", {"mcp_connect_flow_wrong": "garbage"}, 400, "invalid_request"), + (None, None, 401, "login_required"), + ("u2", None, 403, "access_denied"), + ], +) +async def test_describe_connect_flow_refuses_exactly_like_the_finish_step( + session_user_id, cookies, expected_status, expected_error +): + """The page's read of the flow is gated the same way minting is: the HttpOnly cookie for + that handle must open and the signed-in user must be the sealed one. A lure link with a + made-up handle therefore learns nothing and starts nothing.""" + client_id = (await _register([REDIRECT_URI]))["client_id"] + response = _authorize(client_id, session_user_id="u1") + described = await _describe_page(response, session_user_id=session_user_id, cookies=cookies) + assert described.status_code == expected_status + assert json.loads(described.body)["error"] == expected_error + + @pytest.mark.asyncio async def test_token_rejects_resource_conflicting_with_sealed_scope(): """RFC 8707 section 2.2: redeeming a scoped code (or rotating a scoped refresh token) @@ -1005,7 +1213,7 @@ async def test_token_rejects_resource_conflicting_with_sealed_scope(): with patch(_MANAGER_PATCH) as manager: manager.get_mcp_server_by_name.return_value = github response = _scoped_authorize(client_id, SCOPED_RESOURCE) - code = await _finish_connect_page(response) + code = await _finish_connect_page(response, scoped_server=github) with patch(_MANAGER_PATCH) as manager: manager.get_mcp_server_by_name.return_value = linear diff --git a/ui/litellm-dashboard/src/app/chat/integrations/page.tsx b/ui/litellm-dashboard/src/app/chat/integrations/page.tsx index 30ce62d8081..a663e16c2b2 100644 --- a/ui/litellm-dashboard/src/app/chat/integrations/page.tsx +++ b/ui/litellm-dashboard/src/app/chat/integrations/page.tsx @@ -1,43 +1,19 @@ "use client"; -import { Suspense, useEffect } from "react"; -import { useRouter, useSearchParams } from "next/navigation"; +import { Suspense } from "react"; import { useChatShell } from "@/contexts/ChatShellContext"; -import MCPAppsPanel from "@/components/chat/MCPAppsPanel"; -import ConnectFlowBanner from "@/components/chat/ConnectFlowBanner"; +import ConnectFlowSurface from "@/components/chat/ConnectFlowSurface"; // useSearchParams() requires a Suspense boundary for static export. function IntegrationsPageContent() { const { accessToken, selectedMCPServers, setSelectedMCPServers } = useChatShell(); - const router = useRouter(); - const searchParams = useSearchParams(); - const oauthReturn = searchParams.get("mcpOauthReturn"); - // Set by the gateway DCR authorize when a DCR client sends the user here to - // authorize servers before finishing sign-in (see gateway_dcr_flow.py). The - // handle keys the sealed per-flow cookie; connect_client is the client origin - // for display only. connect_flow is NOT cleaned from the URL: the finish form - // needs it, and the sealed cookie (not the URL) is the security boundary. - const connectFlow = searchParams.get("connect_flow"); - const connectClient = searchParams.get("connect_client"); - - // Clean up the OAuth return param after it's been consumed — real routing means - // we no longer need it to pick a tab, but it should not linger in the address bar. - useEffect(() => { - if (oauthReturn) { - const url = new URL(window.location.href); - url.searchParams.delete("mcpOauthReturn"); - router.replace(url.pathname + url.search); - } - }, [oauthReturn, router]); return (
Connect your MCP servers to {clientLabel}
-- Authorize the servers you want to use below, then click Finish connecting to return to {clientLabel}. -
+{copy[0]}
+{copy[1]}