Pr 330

tests/pytest/datasets/klavis_mcp_test.jsonl

@@ -0,0 +1,15 @@
+{"messages": [ { "role": "system", "content": "You are a helpful assistant that can answer questions about Gmail. You have access to Gmail to help you find information." }, { "role": "user", "content": "Fetch the first 5 emails in my gmail inbox and get the sender. Output the sender only." } ], "ground_truth": "The response contains 5 email addresses." }
+{"messages": [ { "role": "system", "content": "You are a helpful assistant that can answer questions about Notion. You have access to tools to Notion you find information." }, { "role": "user", "content": "How many notion pages are in MCPMark Source Hub?" } ], "ground_truth": "10" }
+{"messages": [ { "role": "system", "content": "You are a helpful assistant that can answer questions about Notion. You have access to tools to Notion you find information." }, { "role": "user", "content": "In the notion Toronto guide, help me to find a pizza restaurant which is able to takeout." } ], "ground_truth": "Pizzeria Badiali" }
+{"messages": [ { "role": "system", "content": "You are a helpful assistant that can answer questions about Notion. You have access to tools to Notion you find information." }, { "role": "user", "content": "Check Japan Travel Planner page, help me to calculate how much did I spend in accomondation." } ], "ground_truth": "$373.63" }
+{"messages": [ { "role": "system", "content": "You are a helpful assistant that can answer questions about Notion. You have access to tools to Notion you find information." }, { "role": "user", "content": "Check Japan Travel Planner page, help me to check how many tokyo attractions I've visited." } ], "ground_truth": "2" }
+{"messages": [ { "role": "system", "content": "You are a helpful assistant that can answer questions about Notion. You have access to tools to Notion you find information." }, { "role": "user", "content": "Check Company In A Box page, help me to find how many presses did we have during 2018. You can find the presses in company wiki." } ], "ground_truth": "3" }
+{"messages": [ { "role": "system", "content": "You are a helpful assistant that can answer questions about Notion. You have access to tools to Notion you find information." }, { "role": "user", "content": "Check Company In A Box page, help me to figure out how many FAQ items under training & upskilling category."} ], "ground_truth": "4" }
+{"messages": [ { "role": "system", "content": "You are a helpful assistant that can answer questions about Outlook Calendar. You have access to Outlook Calendar to help you find information." }, { "role": "user", "content": "How many events do I have today?" } ], "ground_truth": "3" }
+{"messages": [ { "role": "system", "content": "You are a helpful assistant that can answer questions about Outlook Calendar. You have access to Outlook Calendar to help you find information." }, { "role": "user", "content": "How many events do I have during business days this week?" } ], "ground_truth": "8" }
+{"messages": [ { "role": "system", "content": "You are a helpful assistant that can answer questions about Outlook Calendar. You have access to Outlook Calendar to help you find information." }, { "role": "user", "content": "How many available time(in hour) do I have during my business hour the next working day?" } ], "ground_truth": "2 hour" }
+{"messages": [ { "role": "system", "content": "You are a helpful assistant that can answer questions about Outlook Calendar. You have access to Outlook Calendar to help you find information." }, { "role": "user", "content": "How many events do I have on Oct 15 2025?" } ], "ground_truth": "4" }
+{"messages": [ { "role": "system", "content": "You are a helpful assistant that can answer questions about Outlook Calendar. You have access to Outlook Calendar to help you find information." }, { "role": "user", "content": "How many events do I have during business days of the week of Oct 15 2025?" } ], "ground_truth": "9" }
+{"messages": [ { "role": "system", "content": "You are a helpful assistant that can answer questions about Outlook Calendar. You have access to Outlook Calendar to help you find information." }, { "role": "user", "content": "How many events do I have on next week's Thursday?" } ], "ground_truth": "2" }
+{"messages": [ { "role": "system", "content": "You are a helpful assistant that can answer questions about Outlook Calendar. You have access to Outlook Calendar to help you find information." }, { "role": "user", "content": "How many events do I have on next week's buisiness day?" } ], "ground_truth": "5" }
+

tests/pytest/mcp_configurations/klavis_strata_mcp.json

@@ -1,8 +1,8 @@
 {
   "mcpServers": {
     "klavis-strata": {
-      "url": "https://strata.klavis.ai/mcp/",
-      "authorization": "Bearer ${KLAVIS_API_KEY}"
+      "url": "https://gmail-mcp-server.klavis.ai/mcp/?instance_id=74ddb651-9204-480e-9fe4-4a90d9f10c69",
+      "authorization": "Bearer ${KLAVIS_AUTH_TOKEN}"
     }
   }
 }

tests/pytest/test_pytest_klavis_mcp.py

@@ -13,10 +13,16 @@ class ResponseFormat(BaseModel):
     score: float
 
 
+"""
+You should copy https://painted-tennis-ebc.notion.site/MCPMark-Source-Hub-23181626b6d7805fb3a7d59c63033819
+into your Notion for the notion test.
+"""
+
+
 @evaluation_test(
-    input_dataset=["tests/pytest/datasets/gmail_inbox.jsonl"],
+    input_dataset=["tests/pytest/datasets/klavis_mcp_test.jsonl"],
     rollout_processor=AgentRolloutProcessor(),
-    completion_params=[{"model": "fireworks_ai/accounts/fireworks/models/kimi-k2-instruct-0905"}],
+    completion_params=[{"model": "fireworks_ai/accounts/fireworks/models/kimi-k2-thinking"}],
     mode="pointwise",
     mcp_config_path="tests/pytest/mcp_configurations/klavis_strata_mcp.json",
 )
@@ -28,7 +34,7 @@ async def test_pytest_klavis_mcp(row: EvaluationRow) -> EvaluationRow:
         api_key=os.environ["FIREWORKS_API_KEY"], base_url="https://api.fireworks.ai/inference/v1"
     ) as client:
         response = await client.chat.completions.create(
-            model="accounts/fireworks/models/kimi-k2-instruct-0905",
+            model="accounts/fireworks/models/kimi-k2-thinking",
             messages=[
                 {
                     "role": "system",
@@ -47,6 +53,7 @@ async def test_pytest_klavis_mcp(row: EvaluationRow) -> EvaluationRow:
         response_text = response.choices[0].message.content
         logger.info("response_text: %s", response_text)
         score = json.loads(response_text or "{}")["score"]
+
         row.evaluation_result = EvaluateResult(
             score=score,
             reason=response_text,