Extract keywords from Excel news data without coding: Bareun MCP + an AI coding tool

Sometimes you want to know what appears most often (the key keywords) in thousands or tens of thousands of sentences collected in Excel. Most Korean keywords are nouns, so the most robust approach is to split sentences into morphemes, pick out only the nouns accurately, and count them. The Bareun morphological analyzer does this for you.

This document is a guide for people who know nothing about programming. It lets you follow along step by step to finish keyword analysis of an Excel file by just telling an AI coding tool (Claude Code or Codex) what you want, without writing a single line of code yourself.

What you get by following this document

Feed it an Excel file with more than 10,000 sentences, like one downloaded from the Korea Press Foundation's BigKinds, and → a ranked table of frequent keywords (nouns) and a bar chart are created in a new Excel file. The AI writes and runs the code for you. You only make requests in plain language.

Setup time

You install and set up three things only once (① Python ② an AI coding tool ③ the Bareun connection). This usually takes 20 to 30 minutes. After that, you can analyze a different Excel file in a few minutes.

The big picture: what do you install, and why?

You prepare three things. Each has its own role.

What you need	Its role in one line	Analogy
Python	The engine that actually runs the analysis code the AI creates	The kitchen where you cook
AI coding tool (Claude Code, Codex)	An assistant that understands your words and writes and runs code for you	The cook working in the kitchen
Bareun MCP server	A specialist tool that accurately analyzes Korean sentences into morphemes and picks out nouns	The knife and scale the cook uses

graph LR
  YOU[Person: asks in words] --> AI[AI coding tool<br/>Claude Code, Codex];
  AI -->|Write and run code| PY[Python];
  AI -->|Korean morphological analysis| BAREUN[Bareun MCP server];
  PY --> XLSX[Read and aggregate Excel];
  BAREUN --> NOUN[Extract nouns];
  XLSX --> OUT[Keyword ranking + chart<br/>new Excel file];
  NOUN --> OUT;

Why do you need morphological analysis?

In Korean, a particle (josa) attaches to a noun to form one chunk, as in 정부가, 정부는, and 정부를. If you count by word spacing alone, these three are caught as different words, and the keyword statistics are thrown off. With Bareun's morphological analysis, particles and endings are separated and merged into the single word 정부, so the count is accurate. For more background, see Keyword extraction and text mining.

Checklist of what you need

A PC with internet access (Windows or macOS)
A Bareun API key (issued in Step 0 below; a string starting with koba-)
The Excel file to analyze (for example, news data downloaded from BigKinds)
An account for an AI coding tool: either Claude Code (Anthropic account) or Codex (ChatGPT account)

Should I use Claude Code or Codex?

You need only one of the two. Choose the one that matches the subscription you already use (Claude or ChatGPT). This document guides you through both tools separately. Click your tool on the tabs (Claude Code / Codex) on the screen and follow along.

Step 0: Get a Bareun API key

The Bareun MCP server identifies users with an API key. You need a key to run analysis.

In a web browser, go to bareun.ai and sign up.
Complete email verification from the email you receive after signing up.
After you log in, check your API key on the "Login – My info (My information)" screen. The key is a long string starting with koba-.
Copy this key. You will paste it as is in place of 발급받은_API_키 ("your issued API key") later.

Treat your API key like a password

Keep your API key hidden from others. Be careful not to expose it in messengers, screen sharing, and so on. For pricing, usage, and issuance, see Using the cloud.

Step 1: Install Python

Python is the engine that actually runs the analysis code the AI creates. You will not handle Python yourself. You only need to install it.

On python.org/downloads, click the yellow "Download Python" button to download the installer.
Run the downloaded file. On the first installer screen, be sure to turn on the Add python.exe to PATH (or Add Python to PATH) checkbox at the very bottom. Then click Install Now.
When the installation finishes, open PowerShell (search for "PowerShell" in the Start menu) and enter the following.
```
python --version
```
If a version such as Python 3.xx.x appears, it worked.

On python.org/downloads, download the macOS installer (.pkg), run it, and click Continue → Install as prompted.
When the installation finishes, open Terminal (search for "Terminal" in Spotlight) and enter the following.
```
python3 --version
```
If a version such as Python 3.xx.x appears, it worked.

New to the terminal?

The terminal (PowerShell on Windows, Terminal on macOS) is a black window where you type commands as text. Just copy the command in the gray box in this document, paste it as is, and press Enter. To paste, you usually right-click or press Ctrl+V (⌘+V on a Mac).

Step 2: Install an AI coding tool

You install the assistant that understands your words and writes code for you. You need to install only one of Claude Code or Codex.

Paste the following single line into the terminal (PowerShell on Windows, Terminal on macOS) and press Enter. Unlike Python, this installation method does not need anything like Node.js separately.

irm https://claude.ai/install.ps1 | iex

curl -fsSL https://claude.ai/install.sh | bash

When the installation finishes, enter the following to run it, and log in through the browser as the screen instructs (you need a paid Anthropic account such as Claude Pro or Max).

claude

To check that the installation worked, enter claude --version, and if there is a problem, enter claude doctor. For detailed installation instructions, see the official Claude Code setup documentation.

Codex is a terminal-based AI coding tool that you use with a ChatGPT account. Paste the following line and press Enter.

powershell -ExecutionPolicy ByPass -c "irm https://chatgpt.com/codex/install.ps1 | iex"

curl -fsSL https://chatgpt.com/codex/install.sh | sh

When the installation finishes, enter the following to run it, and log in with your ChatGPT account as prompted.

codex

For detailed installation instructions, see the official Codex quickstart.

If you see 'command not found' after installing

Close the terminal completely, reopen it, and run the same command once more. Sometimes a restart is needed for a new window to recognize where the installed program is located.

Step 3: Connect the Bareun MCP server

Now you hand the assistant (the AI) a specialist tool called Bareun. Once this connection is in place, the AI calls Bareun directly when it analyzes Korean sentences and performs accurate morphological analysis. (A detailed description of the Bareun MCP server is in Using Bareun as an MCP server.)

In the commands below, you must replace 발급받은_API_키 ("your issued API key") with your actual key (koba-...) that you copied in Step 0.

Paste the following single line into the terminal (changing only the key) and press Enter.

claude mcp add --transport http bareun https://api.bareun.ai:443/mcp --header "api-key: 발급받은_API_키"

Check that it was registered correctly.

claude mcp list

If bareun appears in the list, it worked. During a Claude Code conversation, you can also enter /mcp to see the tools Bareun provides (such as analyze_syntax).

Codex registers servers in a configuration file (config.toml). The file is located as follows.

Windows: C:\Users\username\.codex\config.toml
macOS: ~/.codex/config.toml (the .codex folder in your home folder)

Open this file in Notepad (or a text editor), add the following at the very bottom, and save it. (Change only 발급받은_API_키.)

[mcp_servers.bareun]
url = "https://api.bareun.ai:443/mcp"
http_headers = { "api-key" = "발급받은_API_키" }

If finding or creating the file is hard

With Codex running, you can also ask in words like this: "Add the Bareun MCP server to my ~/.codex/config.toml file. The url is https://api.bareun.ai:443/mcp, and the api-key in http_headers is koba-...." Codex creates the file and fills in the contents. After saving, quit and restart Codex once for it to take effect.

Bareun MCP is provided by Bareun builds that include the Korean spell checker

The cloud address above (api.bareun.ai) is a Bareun build that includes the spell checker, so you can use MCP right away. If you connect a Bareun server installed inside your company, change the address to that server's address and check that it is a Bareun build that includes the Korean spell checker (a Bareun build for morphological analysis only has no /mcp).

Step 4: Prepare the Excel file to analyze

When you choose "Excel download" in BigKinds, you usually get a file with the following columns. Of these, the ones used for analysis are mainly the 제목 (title) and 본문 (body) columns (the cells that hold the actual sentences).

Common column name	Content	Used for analysis?
`일자` (date)	Article date (for example, `20260105`)	When splitting by period
`언론사` (news outlet)	Outlet name	When splitting by outlet
`제목` (title)	Article title sentence	✅ Target for keyword extraction
`본문` (body)	Article body sentences	✅ Target for keyword extraction
`키워드` (keywords)	Words BigKinds extracted in advance	For reference
`URL`	Article address	Usually not used

Practice with a sample file first

We prepared a small sample Excel file with the same shape as the real file. If you do not have your own file or the structure is confusing, follow along with this first. → Download the sample Excel file (bigkinds-sample.xlsx)

This sample contains 12 news items and has the same columns as the table above. Your real file (with more than 10,000 sentences) is analyzed the same way as long as the column structure is similar.

Put the file in one folder. For example, create a 키워드분석 ("keyword analysis") folder on your desktop and put the Excel file (for example, news.xlsx) in it. Then, if you run the AI coding tool inside that folder, the AI can find the file right away, which is convenient.

cd $HOME\Desktop\키워드분석
claude

(If you use Codex, change the last line to codex.)

cd ~/Desktop/키워드분석
claude

(If you use Codex, change the last line to codex.)

If moving into the folder is hard

You can also open that folder in Windows Explorer (or macOS Finder), run the AI tool, and tell it the path in words, such as "I want to analyze news.xlsx in the 키워드분석 folder on my desktop".

Step 5: Ask for the analysis in words (copy-and-paste prompt)

This is the key step. With the AI coding tool running, copy the request below as is, paste it, and press Enter. You only need to change the file name (news.xlsx) and the column name (본문) to match yours.

Request to copy and paste

Analyze the Excel file news.xlsx in this folder.

Use the sentences in the 본문 column (or the 제목 column if there is no 본문).
Using the morphological analysis of the Bareun (bareun) MCP, extract only the nouns (common nouns NNG and proper nouns NNP) from each sentence. Exclude one-character nouns and weak dependent nouns (NNB).
Add up the number of occurrences of the extracted nouns and rank them by frequency.
Save the result as a new Excel file named 키워드결과.xlsx, with two columns: 키워드 (keyword) and 빈도 (frequency).
Also draw a bar chart of the top 20 and save it as an image file.

There are a lot of sentences (thousands to tens of thousands), so don't query one at a time. Proceed with an efficient way to process them all at once. If you need a Python package along the way, feel free to install it.

After the AI explains its plan, it writes and runs code on its own that reads the Excel file → extracts nouns with Bareun → counts frequencies → creates the result file. If it asks "May I run this command?" or "May I use this tool?" along the way, choose Yes/Allow.

Claude Code asks for permission before reading files or running commands. Press Yes to proceed. It may also ask for permission when it uses a Bareun tool (analyze_syntax).

I have more than 10,000 sentences. Is that okay? The two ways of using Bareun

For small amounts (tens to hundreds of sentences), the AI calls the Bareun MCP tool directly and analyzes right away. For thousands to tens of thousands of sentences, like your file, the AI creates a short Python program (the Bareun Python library bareunpy) that uses the same Bareun service and the same API key, and processes everything quickly in one go. Both use the same Bareun, so analysis quality is the same, and you do not need to touch any code either way. The last paragraph of the request above tells the AI to choose this efficient approach.

Step 6: See the results

When the AI finishes, 키워드결과.xlsx (and the bar chart image) appears in the folder. When you open the Excel file, you see a ranking of frequent keywords like the following (this is how it looks when run on the sample file).

키워드 (keyword)	빈도 (frequency)
인공지능 (artificial intelligence)	9
반도체 (semiconductor)	6
수출 (export)	6
정부 (government)	3
교육 (education)	2
…	…

With particles and endings removed, key terms such as 인공지능 and 반도체 are clearly gathered. If you do not like the result, you can just ask again in words to change it, for example: "Exclude entries that are only numbers", "Extend it to the top 50", "Also save it as CSV, not just Excel".

Going further: a collection of follow-up requests

In the same way, you can take the analysis deeper by following up with the requests below.

What you want to do	Say this
See only names of people, organizations, and places	"Extract only proper nouns (NNP) and make a separate ranking of frequently mentioned people and organizations."
See the monthly trend	"Group the `일자` column by month and make a table showing how the top 10 keywords change each month."
Compare by news outlet	"Split by the `언론사` column and compare the keywords each outlet uses most often."
Remove stopwords	"Make a list of meaningless words such as '기자·오늘·관련' (reporter, today, related) and remove them from the results."
Word cloud	"Also make a word cloud image from the top keywords."
Two characters or more only	"Exclude one-character nouns and count only nouns of two or more characters."

Why proper nouns (NNP) are a strength

Bareun was trained on a large corpus that includes 100 million eojeol (space-delimited word units) of BigKinds news, so it is strong at recognizing current-affairs proper nouns such as names of people and organizations. It is especially well suited to analyzing the people and organizations in news data. For detailed cases, see Extracting news keywords.

When it does not work (troubleshooting)

I entered `python` and it says "not found"

The most common cause is that you missed the Add Python to PATH checkbox during installation. Reinstall Python and be sure to turn that checkbox on. If it still does not work, close the terminal and reopen it. On macOS, you may need to enter python3 instead of python.

Bareun (bareun) does not show up in the AI tool

Claude Code: Check that bareun is in claude mcp list, and that the tool appears with /mcp during a conversation. Check that /mcp is at the end of the address and that you put your real key in the api-key header.
Codex: Check that you restarted Codex after saving config.toml. Also check the address and key for typos.

I get a message like "no permission / key rejected (401)"

Either the API key is wrong, or spaces or quotation marks were wrongly included before or after koba-. Copy the key again from Step 0 and paste it exactly. For key issuance and usage, see Using the cloud.

There are too many sentences and it takes a long time

With tens of thousands of rows it can take several minutes or more. If you tell the AI "First run just 100 rows as a test and show me the shape of the result, then run the whole thing if it looks fine", you can proceed and check safely.

Strange words are mixed into the result keywords

If words like 기자 (reporter), 것 (thing), and 수 (way, number) bother you, give the AI a list and say "Remove these words as stopwords." Asking it to exclude dependent nouns (NNB) or one-character nouns also makes the result cleaner.

Frequently asked questions

Q. Do I really not have to write a single line of code?

Right. The AI coding tool writes and runs the code. You only say in plain language what you want and click to allow execution. If you do not like the result, you can ask in words to change it.

Q. Why do I install Python? Do I have to learn Python?

No. Python is just the engine that runs the analysis code the AI creates. You do not need to learn it or handle it yourself. Once installed, you will not need to think about it again.

Q. Does the Bareun MCP alone analyze all 10,000 rows?

For small amounts, the AI calls the Bareun MCP tool directly. For thousands to tens of thousands of rows, the AI creates a short Python program (bareunpy) that uses the same Bareun service and the same API key and processes everything quickly in one go. Either way, analysis quality is the same Bareun, and you never need to touch the code.

Q. Which is better, Claude Code or Codex?

Both do this job well. Pick just one to match the subscription you already use (Claude or ChatGPT). Only the registration method differs. How you hand off the analysis (asking in words) is the same.

Q. Can it be a CSV or text file instead of Excel?

Yes. Change only the file name in the request and tell the AI "it's a CSV file", and it will read it accordingly. You can also ask it to save the result in the form you want, such as Excel, CSV, or an image.

Q. What if nouns in the analysis result are caught wrongly?

For new names of people and organizations that appeared after training, register them in the custom dictionary and Bareun will recognize them as a single word. You can also ask the AI to register custom dictionary entries for you.

Using Bareun as an MCP server: the full guide to the Bareun MCP server's tools, authentication, and registration
Keyword extraction and text mining: why noun frequency makes robust keywords
Extracting news keywords: analysis centered on proper nouns such as names of people and organizations
Using the cloud: API key issuance and usage
POS tag table: the meaning of POS codes such as NNG, NNP, and NNB