chore: bump auto_page_size_threshold to 100 (#253)

* chore: bump auto_page_size_threshold from 50 to 100

Skip the page_size prompt for datasets up to 100 rows instead of 50,
reducing unnecessary user interactions for small datasets.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* fix: replace hardcoded 50 references with settings.auto_page_size_threshold

- models.py: page_size default now uses the config value
- tool_helpers.py: agent prompt uses dynamic threshold instead of literal "50"
- app.py: instructions use f-string with threshold setting
- test comment: removed hardcoded value

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>

Author: RafaelPo

everyrow-mcp/src/everyrow_mcp/app.py

@@ -85,7 +85,7 @@ async def no_auth_http_lifespan(_server: FastMCP):
         )
 
 
-_INSTRUCTIONS_COMMON = """\
+_INSTRUCTIONS_COMMON = f"""\
 You are connected to the everyrow MCP server. everyrow dispatches web research \
 agents that search the internet, read pages, and return structured results for \
 every row in a dataset.
@@ -103,7 +103,7 @@ async def no_auth_http_lifespan(_server: FastMCP):
 ## Key rules
 - If a session_url appears in the submission response, share it with the user. If none is present, do not mention it.
 - Never guess or fabricate results — always wait for the task to complete.
-- For small datasets (< 50 rows), prefer passing `data` directly.
+- For small datasets (<= {settings.auto_page_size_threshold} rows), prefer passing `data` directly.
 - For larger datasets, use `everyrow_upload_data` to get an artifact_id first.
 """
 

everyrow-mcp/src/everyrow_mcp/config.py

@@ -89,7 +89,7 @@ class Settings(BaseSettings):
         description="Maximum rows allowed in inline data (list[dict]).",
     )
     auto_page_size_threshold: int = Field(
-        default=50,
+        default=100,
         description="If total rows <= this value, skip asking the user for page_size and load all rows directly.",
     )
 

everyrow-mcp/src/everyrow_mcp/models.py

@@ -707,7 +707,7 @@ def validate_task_id(cls, v: str) -> str:
         ge=0,
     )
     page_size: int = Field(
-        default=50,
+        default=settings.auto_page_size_threshold,
         description=(
             "Number of result rows to load into your context so you can read them. "
             "The user has access to all rows via the widget regardless of this value. "

everyrow-mcp/src/everyrow_mcp/tool_helpers.py

@@ -446,7 +446,7 @@ def progress_message(self, task_id: str) -> str:
                     else:
                         next_call = dedent(f"""\
                             IMPORTANT: Do NOT call everyrow_results yet.\
-                             First, ask the user: "The task produced {self.total} rows. How many would you like me to load into my context so I can read them? (default: 50). You will have access to all of them via the widget.".\
+                             First, ask the user: "The task produced {self.total} rows. How many would you like me to load into my context so I can read them? (default: {settings.auto_page_size_threshold}). You will have access to all of them via the widget.".\
                              The answer the user provides will correspond to the `page_size`.\
                              After the user responds, call everyrow_results(task_id='{task_id}', page_size=N).""")
                 else:

everyrow-mcp/tests/test_stdio_content.py

@@ -703,7 +703,7 @@ async def test_progress_http_completed_no_output_path_hint(self):
         human_text = result[-1].text
         assert "output_path" not in human_text
         assert "everyrow_results" in human_text
-        # total=5 is below auto_page_size_threshold (50), so the model
+        # total=5 is below auto_page_size_threshold, so the model
         # should be told to load all rows directly instead of asking.
         assert "load all rows" in human_text.lower()