Rank Data by External Metrics

Add the everyrow connector if you haven't already. Then upload a CSV of the top 300 PyPI packages (with package and monthly_downloads columns) and ask Claude:

Rank these packages by days since their last release. Look up each package on pypi.org to find the release date. Sort by most recently released first.

All 300 packages researched in about 6.5 minutes. Results range from packages released today to ones untouched for 8+ years.

Claude Code's web search works well for looking up a few packages. When you need two separate metrics researched for each of 300 packages, that is hundreds of individual lookups that need to happen in parallel.

Here, we get Claude Code to rank 300 PyPI packages by two metrics that require external lookup: days since last release (from PyPI) and number of contributors (from GitHub).

Add everyrow to Claude Code if you haven't already:

claude mcp add futuresearch --scope project --transport http https://mcp.futuresearch.ai/mcp

The dataset is the top 300 PyPI packages by monthly downloads, fetched from the top-pypi-packages API. The only columns are package and monthly_downloads. No release dates, no contributor counts. Tell Claude:

Rank these 300 PyPI packages by days since their last release.
Look up each package on pypi.org to find the release date.
Sort by most recently released first.

Claude calls everyrow's rank MCP tool, then polls for progress until the operation completes:

Tool: everyrow_rank
├─ task: "Find the number of days since this package's last release on PyPI..."
├─ input_csv: "/Users/you/top_pypi_packages.csv"
├─ field_name: "days_since_release"
├─ field_type: "int"
└─ ascending_order: true

→ Submitted: 300 rows for ranking.
  Session: https://futuresearch.ai/sessions/7a461cd9-056b-42b2-b335-8d52fe3f685c
  Task ID: 7a46...

Tool: everyrow_progress
├─ task_id: "7a46..."
→ Running: 0/300 complete, 300 running (15s elapsed)

Tool: everyrow_progress
→ Running: 150/300 complete, 150 running (120s elapsed)

...

Tool: everyrow_progress
→ Completed: 300/300 (0 failed) in 236s.

Tool: everyrow_results
├─ task_id: "7a46..."
├─ output_path: "/Users/you/pypi_ranked_by_release.csv"
→ Saved 300 rows to /Users/you/pypi_ranked_by_release.csv

Under the hood, everyrow dispatched LLM-powered web research agents to look up each package on PyPI. The agents found that fastapi was released today, while webencodings hasn't been updated in 8.9 years.

The same approach works for any metric you can describe. A second rank call on the same data, asking for number of GitHub contributors, ran in parallel:

Tool: everyrow_rank
├─ task: "Find the number of contributors to this package's GitHub repository..."
├─ input_csv: "/Users/you/top_pypi_packages.csv"
├─ field_name: "num_contributors"
├─ field_type: "int"
└─ ascending_order: false

→ Completed: 300/300 in 391s.

Both operations completed in ~6.5 minutes of wall clock time. View the sessions.

The everyrow Python SDK can rank or sort data on criteria you don't have in your dataset, if it can find them on the web. It dispatches LLM-powered web research agents to look up each row in parallel.

This guide shows how to rank 300 PyPI packages by two different metrics that require external lookup: days since last release (from PyPI) and number of contributors (from GitHub).

pip install everyrow
export EVERYROW_API_KEY=your_key_here  # Get one at futuresearch.ai/api-key

The dataset is the top 300 PyPI packages by monthly downloads, fetched from the top-pypi-packages API. The only columns are package and monthly_downloads—no release dates.

import asyncio
import requests
import pandas as pd
from everyrow.ops import rank

# Fetch top PyPI packages
response = requests.get(
    "https://hugovk.github.io/top-pypi-packages/top-pypi-packages-30-days.min.json"
)
packages = response.json()["rows"][50:350]  # Skip AWS libs at top
df = pd.DataFrame(packages).rename(
    columns={"project": "package", "download_count": "monthly_downloads"}
)

async def main():
    result = await rank(
        task="""
            Find the number of days since this package's last release on PyPI.
            Look up the package on pypi.org to find the release date.
            Return the number of days as an integer.
        """,
        input=df,
        field_name="days_since_release",
        field_type="int",
        ascending_order=True,  # Most recent first
    )
    print(result.data[["package", "days_since_release"]])

asyncio.run(main())

                   package  days_since_release
0                pyparsing                   0
1                 httplib2                   1
2              yandexcloud                   2
3             multiprocess                   2
4                  pyarrow                   3
...
295        ptyprocess                1850
296              toml                1907
297               ply                2897
298      webencodings                3213

The SDK dispatched LLM-powered web research agents to review each row. It found that pyparsing was released today (Jan 20 2026), and webencodings hasn't been updated in 8.8 years.

The same approach works for any metric you can describe. Here's the same dataset ranked by number of GitHub contributors:

result = await rank(
    task="""
        Find the number of contributors to this package's GitHub repository.
        Look up the package's source repo from PyPI, then find the contributor
        count on GitHub. Return the number as an integer.
    """,
    input=df,
    field_name="num_contributors",
    field_type="int",
    ascending_order=False,  # Most contributors first
)

                    package  num_contributors
0                     torch              4191
1                 langchain              3858
2            langchain-core              3858
3              transformers              3608
4              scikit-learn              3157
...
295        jsonpath-ng                 2
296         et-xmlfile                 1
297     beautifulsoup4                 1
298        ruamel-yaml                 1
299            pkginfo                 1

torch has 4,191 contributors; pkginfo has 1. The task prompt tells the agent what to look up and where—citation counts, benchmark scores, API response times, or anything else you can describe.