feat(everyrow-mcp): wire include_partial_rows and improve progress prompt (#4946)

## Summary
- Wire `include_partial_rows` (default `false`) into `everyrow_progress`
so callers can opt in to receiving newly completed row data alongside
status counts and agent summaries
- Improve the progress message prompt to ask the LLM for meaningful
highlights (interesting findings, patterns, notable values) rather than
a generic "briefly comment"
- Add proper `Field()` descriptor with description for
`include_partial_rows` on `ProgressInput`

## Test plan
- [x] All 422 existing MCP server tests pass
- [ ] Manual test: call `everyrow_progress` without
`include_partial_rows` — should return only status + summaries, no row
data
- [ ] Manual test: call `everyrow_progress(include_partial_rows=true)` —
should include newly completed rows
- [ ] Verify LLM produces more meaningful progress updates with the new
prompt

🤖 Generated with [Claude Code](https://claude.com/claude-code)

---------

Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Sourced from commit b01635969cd1d9f2c81bcdf5f2a9fc965f0dd548

Author: RafaelPo

everyrow-mcp/src/everyrow_mcp/app.py

@@ -79,21 +79,58 @@ async def no_auth_http_lifespan(_server: FastMCP):
 agents that search the internet, read pages, and return structured results for \
 every row in a dataset.
 
+## Getting data
+
+Most operations need an input dataset. If the user provides a CSV, start from that. \
+Otherwise, help them find one:
+
+1. **Built-in lists** — check `everyrow_browse_lists` first (fast and free). \
+Call with no filters to see all available lists. Many analyses start from one of these.
+2. **URLs** — upload from a URL or Google Sheet via `everyrow_upload_data`.
+3. **From memory** — if you know a good starting list, generate it as inline `data`.
+4. **single_agent** — dispatch a research agent to find or build a list. \
+Works well but slow (3-5 min), so prefer the options above.
+
+## Choosing the right operation
+
+1. **Forecast** — questions about the future. Best prediction accuracy.
+2. **Classify** — binary yes/no or categorical labels (up to ~20 categories). \
+More efficient than open-ended research for categorical answers.
+3. **Rank** — quantitative rating. Prefer an objective metric with units when possible. \
+Use a subjective 0-100 score only if necessary.
+4. **Agent** — open-ended web research when Classify, Rank, and Forecast don't fit. \
+Specify a response schema with descriptive column names (include units, e.g. \
+`population_millions`). Don't add reasoning/justification fields — users can \
+inspect the research behind each row.
+5. **Dedupe / Merge** — data consolidation.
+
 ## Workflow
 1. **Ingest data** — pass `data` (inline list of dicts) or an `artifact_id` \
 (from `everyrow_upload_data` or `everyrow_request_upload_url`) to any processing tool.
-2. **Submit** — call a processing tool (everyrow_agent, everyrow_classify, \
-everyrow_rank, everyrow_dedupe, everyrow_merge, everyrow_forecast). \
-It returns a task_id immediately.
+2. **Submit** — call a processing tool. It returns a task_id immediately.
 3. **Poll** — call `everyrow_progress(task_id)` repeatedly until the task completes. \
-Do NOT add commentary between progress calls — just call again immediately.
+When progress includes new rows or agent activity, give the user a 1-2 sentence \
+status update highlighting interesting findings, then call progress again. \
+When there are no new updates, call progress again immediately without commentary.
 4. **Results** — call `everyrow_results(task_id)` to retrieve the output.
 
+## Session and artifact reuse
+
+Every operation creates a session. After your first operation or upload, pass the \
+returned `session_id` to subsequent operations to keep tasks grouped. When an \
+operation completes, its `artifact_id` can be passed directly to the next operation \
+instead of re-uploading data.
+
 ## Key rules
+- Be concise. Keep summaries to 1-2 sentences. Do not output markdown tables, \
+bullet lists of data rows, JSON, or CSV in chat — the user can see results \
+directly. Only render a table if the user explicitly asks for one.
 - Do not share session URLs with the user unless they explicitly ask for one.
 - Never guess or fabricate results — always wait for the task to complete.
 - 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.
+- After presenting results, mention that the output can be used as input to another \
+operation (e.g. classify then rank, upload then forecast).
 """
 
 _INSTRUCTIONS_STDIO = (

everyrow-mcp/src/everyrow_mcp/models.py

@@ -658,6 +658,11 @@ class ProgressInput(BaseModel):
         "Pass this to only receive new rows and summaries since the last check. "
         "Omit on the first call to see all completed rows so far.",
     )
+    include_partial_rows: bool = Field(
+        default=False,
+        description="Include newly completed rows in the progress response. "
+        "Set to true to receive row data alongside status counts and agent summaries.",
+    )
 
     @field_validator("task_id")
     @classmethod

everyrow-mcp/src/everyrow_mcp/tool_helpers.py

@@ -495,7 +495,7 @@ def progress_message(
         progress_call = f"everyrow_progress(task_id='{task_id}'{cursor_arg})"
 
         if partial_rows or summaries:
-            msg += f"\n\nBriefly comment on these updates for the user, then immediately call {progress_call}."
+            msg += f"\n\nProduce a concise, meaningful update: highlight any interesting findings, patterns, or notable values from the new rows and agent activity above. Then immediately call {progress_call}."
         else:
             msg += f"\nImmediately call {progress_call}."
 

everyrow-mcp/src/everyrow_mcp/tools.py

@@ -1007,7 +1007,7 @@ async def everyrow_progress(
     cursor: str | None = params.cursor
 
     if not ts.is_terminal:
-        if ts.completed > 0:
+        if ts.completed > 0 and params.include_partial_rows:
             (
                 (partial_rows, rows_cursor),
                 (summaries, summary_cursor),