created by Diego Fernandez, maintained by Paradigm LLC

Requires Python 3.9+, gspread 6, and pandas 2 or 3.

Links:

• Documentation
• Source code
• Changelog
• Short video tutorial (from 2017 — the concepts hold, the API has moved on)

A package to easily open an instance of a Google spreadsheet and interact with worksheets through Pandas DataFrames. It enables you to easily pull data from Google spreadsheets into DataFrames as well as push data into spreadsheets from DataFrames. It leverages gspread in the backend for most of the heavylifting, but it has a lot of added functionality to handle things specific to working with DataFrames as well as some extra nice to have features.

The target audience are Data Analysts and Data Scientists, but it can also be used by Data Engineers or anyone trying to automate workflows with Google Sheets and Pandas.

Some key goals/features:

• Be easy to use interactively, with good docstrings and auto-completion
• Nicely handle headers and indexes (including multi-level headers and merged cells)
• Run on Jupyter, headless server, and/or scripts
• Allow storing different user credentials or using Service Accounts
• Automatically handle token refreshes
• Enable handling of frozen rows and columns
• Enable filling in all merged cells when pulling data
• Nicely handle large data sets and auto-retries
• Enable creation of filters
• Handle retries when exceeding 100 second user quota
• When pushing DataFrames with MultiIndex columns, allow merging or flattening headers
• Ability to handle Spreadsheet permissions
• Ability to specify ValueInputOption and ValueRenderOption for specific columns
• Append to a sheet, matching columns to the headers already there
• Find where the table starts in a sheet that opens with a title or a blank row
• Give columns their real dtypes instead of returning everything as strings
• Create a spreadsheet straight into a Drive folder
• Reorder the worksheets in a spreadsheet

Sheets that people actually maintain drift. A column gets renamed between exports, someone adds a title row above the table, every value arrives as text. Three options handle that, and all of them work with no configuration at all.

convert_types gives columns their real dtypes. A column is only converted when every non-empty value in it converts cleanly, so a column that is 99% numbers and one N/A stays as text rather than quietly nulling that row:

df = spread.sheet_to_df(convert_types=True)

detect_layout finds where the table starts, for sheets that open with a report title, a blank row, or a note to the team:

df = spread.sheet_to_df(detect_layout=True)

append adds rows below the data already in a sheet, matching your columns to the headers that are there. Reordered or renamed columns still land in the right place instead of being written positionally:

spread.df_to_sheet(new_rows, append=True)

Each of the three falls back to exact matching, similarity and strict type inference, which covers most sheets on its own. Set an API key and they can also handle renames that don't look alike (revenue to Sales Total), stacked headers, and columns whose type only makes sense from the name:

export GSPREAD_PANDAS_AI_API_KEY=...
export GSPREAD_PANDAS_AI_BASE_URL=https://api.deepseek.com/v1  # default
export GSPREAD_PANDAS_AI_MODEL=deepseek-chat                   # default

Any OpenAI-compatible endpoint works, so DeepSeek, OpenAI, OpenRouter, LiteLLM and a local Ollama are all reachable by changing the base URL. No extra packages are needed.

The model is only ever asked for a small structured answer: a column-to-header mapping, a header row number, a type name from a fixed list. Nothing it returns is executed, and anything naming a column, header or type that isn't really there is discarded. A wrong answer costs you a match, never a wrong value in a cell. Only column names and a bounded sample of rows are ever sent.

To install use pip:

$ pip install gspread-pandas

Or clone the repo:

$ git clone https://github.com/Paradigmllc/gspread-pandas.git
$ pip install .

Before using, you will need to download Google client credentials for your app.

To allow a script to use Google Drive API we need to authenticate our self towards Google. To do so, we need to create a project, describing the tool and generate credentials. Please use your web browser and go to Google console and :

• Choose Create Project in popup menu on the top.
• A dialog box appears, so give your project a name and click on Create button.
• On the left-side menu click on APIs & Services.
• A table of available APIs is shown. Switch Drive API and click on Enable API button. Do the same for Sheets API. Other APIs might be switched off, for our purpose.
• On the left-side menu click on Credentials.
• In section OAuth consent screen select your email address and give your product a name. Then click on Save button.
• In section Credentials click on Add credentials and switch OAuth client ID (if you want to use your own account or enable the use of multiple accounts) or Service account key (if you prefer to have a service account interacting with spreadsheets).
• If you select OAuth client ID:
  • Select Application type item as Desktop app and give it a name.
  • Click on Create button.
  • Click on Download JSON icon on the right side of created OAuth client IDs and store the downloaded file on your file system.
• If you select Service account key
  • Click on Service account dropdown and select New service account
  • Give it a Service account name and ignore the Role dropdown (unless you know you need this for something else, it's not necessary for working with spreadsheets)
  • Note the Service account ID as you might need to give that user permission to interact with your spreadsheets
  • Leave Key type as JSON
  • Click Create and store the downloaded file on your file system.
• Please be aware, the file contains your private credentials, so take care of the file in the same way you care of your private SSH key; Move the downloaded JSON to ~/.config/gspread_pandas/google_secret.json (or you can configure the directory and file name by directly calling gspread_pandas.conf.get_config

Thanks to similar project df2gspread for this great description of how to get the client credentials.

You can read more about it in the configuration docs including how to change the default behavior.

import pandas as pd
from gspread_pandas import Spread, Client

file_name = "http://stats.idre.ucla.edu/stat/data/binary.csv"
df = pd.read_csv(file_name)

# 'Example Spreadsheet' needs to already exist and your user must have access to it
spread = Spread('Example Spreadsheet')
# This will ask to authenticate if you haven't done so before

# Display available worksheets
spread.sheets

# Save DataFrame to worksheet 'New Test Sheet', create it first if it doesn't exist
spread.df_to_sheet(df, index=False, sheet='New Test Sheet', start='A2', replace=True)
spread.update_cells('A1', 'B1', ['Created by:', spread.email])
print(spread)
# <gspread_pandas.client.Spread - User: '<example_user>@gmail.com', Spread: 'Example Spreadsheet', Sheet: 'New Test Sheet'>

# You can now first instanciate a Client separately and query folders and
# instanciate other Spread objects by passing in the Client
client = Client()
# Assumming you have a dir called 'example dir' with sheets in it
available_sheets = client.find_spreadsheet_files_in_folders('example dir')
spreads = []
for sheet in available_sheets.get('example dir', []):
    spreads.append(Spread(sheet['id'], client=client))

IMO, Google sheets is not the right tool for large datasets. However, there's probably good reaons you might have to use it in such cases. When uploading a large DataFrame, you might run into this error.

By default, Spread.df_to_sheet will add rows and/or columns needed to accomodate the DataFrame. Since a new sheet contains a fairly large number of columns, if you're uploading a DF with lots of rows you might exceed the max number of cells in a worksheet even if your data does not. In order to fix this you have 2 options:

1. The easiest is to pass replace=True, which will first resize the worksheet and clear out all values.
2. Another option is to first resize to 1x1 using Spread.sheet.resize(1, 1) and then do df_to_sheet

There's a strange caveat with resizing, so going to 1x1 first is recommended (replace=True already does this). To read more see this issue