feat(agentgateway): make timeout configurable and increase default timeout value (#118)

Signed-off-by: Prashant <prashant.rakheja@sap.com>
Co-authored-by: Nicole Gomes <47161082+NicoleMGomes@users.noreply.github.com>

Author: prashantrakheja

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "sap-cloud-sdk"
-version = "0.18.4"
+version = "0.19.0"
 description = "SAP Cloud SDK for Python"
 readme = "README.md"
 license = "Apache-2.0"

src/sap_cloud_sdk/agentgateway/__init__.py

@@ -53,6 +53,7 @@
 """
 
 from sap_cloud_sdk.agentgateway._models import MCPTool
+from sap_cloud_sdk.agentgateway.config import ClientConfig
 from sap_cloud_sdk.agentgateway.agw_client import create_client, AgentGatewayClient
 from sap_cloud_sdk.agentgateway.exceptions import (
     AgentGatewaySDKError,
@@ -65,6 +66,8 @@
     "create_client",
     # Client class
     "AgentGatewayClient",
+    # Configuration
+    "ClientConfig",
     # Data models
     "MCPTool",
     # Exceptions

src/sap_cloud_sdk/agentgateway/_customer.py

@@ -35,9 +35,6 @@
 # Default credential path for Kyma production deployments
 _CREDENTIALS_DEFAULT_PATH = "/etc/ums/credentials/credentials"
 
-# HTTP timeout for token requests and MCP server calls (seconds)
-_HTTP_TIMEOUT = 30.0
-
 # Resource URN for Agent Gateway token scope (hardcoded - production value)
 _AGW_RESOURCE_URN = "urn:sap:identity:application:provider:name:agent-gateway"
 
@@ -213,6 +210,7 @@ def _create_ssl_context(certificate: str, private_key: str) -> ssl.SSLContext:
 def _request_token_mtls(
     credentials: CustomerCredentials,
     grant_type: str,
+    timeout: float,
     app_tid: str | None = None,
     extra_data: dict | None = None,
 ) -> str:
@@ -255,7 +253,7 @@ def _request_token_mtls(
     try:
         with httpx.Client(
             verify=ssl_context,
-            timeout=_HTTP_TIMEOUT,
+            timeout=timeout,
         ) as client:
             response = client.post(
                 credentials.token_service_url,
@@ -293,6 +291,7 @@ def _request_token_mtls(
 
 def get_system_token_mtls(
     credentials: CustomerCredentials,
+    timeout: float,
     app_tid: str | None = None,
 ) -> str:
     """Get system-scoped token using mTLS client credentials flow.
@@ -301,6 +300,7 @@ def get_system_token_mtls(
 
     Args:
         credentials: Customer credentials.
+        timeout: HTTP timeout in seconds.
         app_tid: BTP Application Tenant ID of subscriber (optional).
 
     Returns:
@@ -310,6 +310,7 @@ def get_system_token_mtls(
     return _request_token_mtls(
         credentials,
         grant_type=_GRANT_TYPE_CLIENT_CREDENTIALS,
+        timeout=timeout,
         app_tid=app_tid,
         extra_data={"response_type": "token"},
     )
@@ -318,6 +319,7 @@ def get_system_token_mtls(
 def exchange_user_token(
     credentials: CustomerCredentials,
     user_token: str,
+    timeout: float,
     app_tid: str | None = None,
 ) -> str:
     """Exchange user token for AGW-scoped token using jwt-bearer grant.
@@ -328,6 +330,7 @@ def exchange_user_token(
     Args:
         credentials: Customer credentials.
         user_token: User's JWT token to exchange.
+        timeout: HTTP timeout in seconds.
         app_tid: BTP Application Tenant ID of subscriber (optional).
 
     Returns:
@@ -337,6 +340,7 @@ def exchange_user_token(
     return _request_token_mtls(
         credentials,
         grant_type=_GRANT_TYPE_JWT_BEARER,
+        timeout=timeout,
         app_tid=app_tid,
         extra_data={
             "assertion": user_token,
@@ -371,6 +375,7 @@ async def _list_server_tools(
     url: str,
     auth_token: str,
     dependency: IntegrationDependency,
+    timeout: float,
 ) -> list[MCPTool]:
     """List tools from a single MCP server.
 
@@ -390,7 +395,7 @@ async def _list_server_tools(
             "Authorization": f"Bearer {auth_token}",
             "x-correlation-id": str(uuid.uuid4()),
         },
-        timeout=_HTTP_TIMEOUT,
+        timeout=timeout,
     ) as http_client:
         async with streamable_http_client(url, http_client=http_client) as (
             read,
@@ -427,6 +432,7 @@ async def _list_server_tools(
 
 async def get_mcp_tools_customer(
     credentials: CustomerCredentials,
+    timeout: float,
     app_tid: str | None = None,
 ) -> list[MCPTool]:
     """List all MCP tools from servers defined in credentials.
@@ -456,7 +462,7 @@ async def get_mcp_tools_customer(
     # Get system token for discovery
     loop = asyncio.get_running_loop()
     system_token = await loop.run_in_executor(
-        None, get_system_token_mtls, credentials, app_tid
+        None, get_system_token_mtls, credentials, timeout, app_tid
     )
 
     tools: list[MCPTool] = []
@@ -471,7 +477,7 @@ async def get_mcp_tools_customer(
         )
 
         try:
-            server_tools = await _list_server_tools(url, system_token, dep)
+            server_tools = await _list_server_tools(url, system_token, dep, timeout)
             tools.extend(server_tools)
             logger.debug("Loaded %d tool(s) from %s", len(server_tools), dep.ord_id)
         except Exception:
@@ -487,6 +493,7 @@ async def call_mcp_tool_customer(
     credentials: CustomerCredentials,
     tool: MCPTool,
     user_token: str | None,
+    timeout: float,
     app_tid: str | None = None,
     **kwargs,
 ) -> str:
@@ -513,7 +520,7 @@ async def call_mcp_tool_customer(
     if user_token:
         # Exchange user token for AGW-scoped token (with principal propagation)
         agw_token = await loop.run_in_executor(
-            None, exchange_user_token, credentials, user_token, app_tid
+            None, exchange_user_token, credentials, user_token, timeout, app_tid
         )
     else:
         # TODO: IBD workaround - use system token when user_token is not available.
@@ -524,15 +531,15 @@ async def call_mcp_tool_customer(
             "Principal propagation will NOT work."
         )
         agw_token = await loop.run_in_executor(
-            None, get_system_token_mtls, credentials, app_tid
+            None, get_system_token_mtls, credentials, timeout, app_tid
         )
 
     async with httpx.AsyncClient(
         headers={
             "Authorization": f"Bearer {agw_token}",
             "x-correlation-id": str(uuid.uuid4()),
         },
-        timeout=_HTTP_TIMEOUT,
+        timeout=timeout,
     ) as http_client:
         async with streamable_http_client(tool.url, http_client=http_client) as (
             read,

src/sap_cloud_sdk/agentgateway/_lob.py

@@ -36,9 +36,6 @@
 
 _DESTINATION_INSTANCE = "default"
 
-# HTTP timeout for MCP server requests (seconds)
-_HTTP_TIMEOUT = 30.0
-
 
 def _ias_dest_name() -> str:
     """Get IAS destination name based on landscape.
@@ -227,7 +224,7 @@ def _fetch_user_auth_sync():
 
 
 async def list_server_tools(
-    dest_url: str, system_auth: str, fragment_name: str
+    dest_url: str, system_auth: str, fragment_name: str, timeout: float
 ) -> list[MCPTool]:
     """List tools from a single MCP server.
 
@@ -241,7 +238,7 @@ async def list_server_tools(
     """
     async with httpx.AsyncClient(
         headers={"Authorization": system_auth, "x-correlation-id": str(uuid.uuid4())},
-        timeout=_HTTP_TIMEOUT,
+        timeout=timeout,
     ) as http_client:
         async with streamable_http_client(dest_url, http_client=http_client) as (
             read,
@@ -273,6 +270,7 @@ async def list_server_tools(
 
 async def get_mcp_tools_lob(
     tenant_subdomain: str,
+    timeout: float,
 ) -> list[MCPTool]:
     """List all MCP tools using LoB flow (destination-based).
 
@@ -309,7 +307,9 @@ async def get_mcp_tools_lob(
 
         try:
             system_auth = await get_system_auth(tenant_subdomain)
-            server_tools = await list_server_tools(mcp_url, system_auth, fragment_name)
+            server_tools = await list_server_tools(
+                mcp_url, system_auth, fragment_name, timeout
+            )
             tools.extend(server_tools)
             logger.debug(
                 "Loaded %d tool(s) from fragment '%s'",
@@ -330,6 +330,7 @@ async def call_mcp_tool_lob(
     tool: MCPTool,
     user_token: str,
     tenant_subdomain: str,
+    timeout: float,
     **kwargs,
 ) -> str:
     """Invoke an MCP tool using LoB flow (destination-based).
@@ -357,7 +358,7 @@ async def call_mcp_tool_lob(
 
     async with httpx.AsyncClient(
         headers={"Authorization": user_auth, "x-correlation-id": str(uuid.uuid4())},
-        timeout=_HTTP_TIMEOUT,
+        timeout=timeout,
     ) as http_client:
         async with streamable_http_client(tool.url, http_client=http_client) as (
             read,

src/sap_cloud_sdk/agentgateway/agw_client.py

@@ -11,6 +11,7 @@
 from typing import Callable
 
 from sap_cloud_sdk.agentgateway._models import MCPTool
+from sap_cloud_sdk.agentgateway.config import ClientConfig
 from sap_cloud_sdk.agentgateway._customer import (
     detect_customer_agent_credentials,
     load_customer_credentials,
@@ -72,15 +73,18 @@ class AgentGatewayClient:
     def __init__(
         self,
         tenant_subdomain: str | Callable[[], str] | None = None,
+        config: ClientConfig | None = None,
     ):
         """Initialize the Agent Gateway client.
 
         Args:
             tenant_subdomain: Tenant subdomain for multi-tenant lookup.
                 Can be a string or a callable returning a string.
                 Required for LoB agents, ignored for Customer agents.
+            config: Client configuration. Uses defaults if not provided.
         """
         self._tenant_subdomain = tenant_subdomain
+        self._config = config or ClientConfig()
 
     @staticmethod
     def _resolve_value(
@@ -153,21 +157,24 @@ async def list_mcp_tools(
                     "Customer agent credentials detected at '%s'", credentials_path
                 )
                 credentials = load_customer_credentials(credentials_path)
-                return await get_mcp_tools_customer(credentials, app_tid)
+                return await get_mcp_tools_customer(
+                    credentials, self._config.timeout, app_tid
+                )
 
             # LoB flow - requires tenant_subdomain
             if app_tid:
                 logger.warning("app_tid parameter ignored for LoB agent flow")
 
             tenant = self._resolve_tenant_subdomain()
-            return await get_mcp_tools_lob(tenant)
+            return await get_mcp_tools_lob(tenant, self._config.timeout)
 
         except AgentGatewaySDKError:
             # Re-raise SDK errors as-is
             raise
         except Exception as e:
             logger.exception("Unexpected error during tool discovery")
-            raise AgentGatewaySDKError(f"Tool discovery failed: {e}") from e
+            cause = _unwrap_exception_group(e)
+            raise AgentGatewaySDKError(f"Tool discovery failed: {cause}") from e
 
     @record_metrics(Module.AGENTGATEWAY, Operation.AGENTGATEWAY_CALL_MCP_TOOL)
     async def call_mcp_tool(
@@ -240,7 +247,12 @@ async def call_mcp_tool(
 
                 credentials = load_customer_credentials(credentials_path)
                 return await call_mcp_tool_customer(
-                    credentials, tool, resolved_user_token, app_tid, **kwargs
+                    credentials,
+                    tool,
+                    resolved_user_token,
+                    self._config.timeout,
+                    app_tid,
+                    **kwargs,
                 )
 
             # LoB flow - requires user_token and tenant_subdomain
@@ -253,20 +265,31 @@ async def call_mcp_tool(
                 logger.warning("app_tid parameter ignored for LoB agent flow")
 
             tenant = self._resolve_tenant_subdomain()
-            return await call_mcp_tool_lob(tool, resolved_user_token, tenant, **kwargs)
+            return await call_mcp_tool_lob(
+                tool, resolved_user_token, tenant, self._config.timeout, **kwargs
+            )
 
         except AgentGatewaySDKError:
             # Re-raise SDK errors as-is
             raise
         except Exception as e:
             logger.exception("Unexpected error during tool invocation")
+            cause = _unwrap_exception_group(e)
             raise AgentGatewaySDKError(
-                f"Tool invocation failed for '{tool.name}': {e}"
+                f"Tool invocation failed for '{tool.name}': {cause}"
             ) from e
 
 
+def _unwrap_exception_group(exc: BaseException) -> BaseException:
+    """Unwrap nested ExceptionGroups to present meaningful error messages."""
+    while isinstance(exc, BaseExceptionGroup) and exc.exceptions:
+        exc = exc.exceptions[0]
+    return exc
+
+
 def create_client(
     tenant_subdomain: str | Callable[[], str] | None = None,
+    config: ClientConfig | None = None,
 ) -> AgentGatewayClient:
     """Create an Agent Gateway client for discovering and invoking MCP tools.
 
@@ -277,6 +300,7 @@ def create_client(
         tenant_subdomain: Tenant subdomain for multi-tenant lookup.
             Can be a string or a callable returning a string.
             Required for LoB agents, ignored for Customer agents.
+        config: Client configuration. Uses defaults if not provided.
 
     Returns:
         AgentGatewayClient instance.
@@ -319,4 +343,4 @@ def create_client(
         )
         ```
     """
-    return AgentGatewayClient(tenant_subdomain=tenant_subdomain)
+    return AgentGatewayClient(tenant_subdomain=tenant_subdomain, config=config)

src/sap_cloud_sdk/agentgateway/config.py

@@ -0,0 +1,17 @@
+"""Configuration for Agent Gateway client."""
+
+from dataclasses import dataclass
+
+DEFAULT_TIMEOUT_SECONDS = 60.0
+
+
+@dataclass
+class ClientConfig:
+    """Configuration options for the Agent Gateway client.
+
+    Attributes:
+        timeout: HTTP timeout in seconds for token requests and MCP server calls.
+            Defaults to 60 seconds.
+    """
+
+    timeout: float = DEFAULT_TIMEOUT_SECONDS