Claude Code Without Coding: Workflows for Non-Programmer Researchers

Who this is for: Researchers who are not programmers — or who code only occasionally — and want to understand what they can actually do with Claude Code before investing time in learning it. This document is a catalogue of real workflows that require zero code, zero scripting, and zero technical expertise beyond launching Claude Code.

The core point: Claude Code is named "Code" because it was designed for software developers. But the underlying capability — an AI that can read and write files on your disk, process many documents in one go, and remember your project context — is just as useful for researchers who never write a line of code.

What "zero coding" means here

You will type natural language instructions. Claude will do the work, including any scripting or commands it needs internally. You do not see, write, or need to understand code. Your interface is plain English.

> Read all PDFs in this folder and extract a list of named persons from each.
  Save to persons.md

Claude reads the files, does the extraction, writes the output. The fact that it may internally use code to parse the PDF is irrelevant to you — the same way you do not need to know how a car engine works to drive.

Workflow 1: Reading and processing documents

Single document — focused read:

> Read this_article.pdf
  What is the main argument? What sources does it rely on?
  Is there anything methodologically unusual about it?

Multiple documents — comparative read:

> Read article_A.pdf and article_B.pdf
  How do their arguments about peasant agency differ?
  Do they cite the same sources?

Folder of documents — batch read:

> Read all PDFs in /seminar_readings/
  For each one: main argument in 2 sentences, key method, one discussion question.
  Output as reading_notes.md

Targeted extraction:

> Read the_report.pdf
  I only care about the funding sources section.
  Find it, summarise it, and quote any specific figures mentioned.

No code. No upload. No copy-paste. See A7.working-with-pdfs for more depth on PDF-specific patterns.

Workflow 2: Writing and editing

Draft revision:

> Read my_draft.docx
  The argument in section 2 is unclear. Revise it for clarity
  without changing the substance. Show me the revised version.

Register and tone check:

> Read this_paragraph.md
  Is the tone appropriate for a peer-reviewed historical journal?
  Flag anything too informal, overclaiming, or hedged to the point of vagueness.

Abstract writing:

> Read my_paper.docx
  Write an abstract of 200 words following this structure:
  background (2 sentences), research question (1 sentence),
  method (1 sentence), main findings (2 sentences), significance (1 sentence).

Grant proposal section:

> I am writing the "Broader Impact" section of an ERC grant proposal.
  The project is about [description].
  Funder priorities from the call: [paste text].
  My draft: [paste or read from file].
  Revise my draft to better align with the funder's language,
  without adding commitments I have not made.

Translating or paraphrasing:

> Read the Latin passage in excerpt.txt
  Give me a close English translation, then a paraphrase in plain modern language.
  Note any terms that are technically difficult to render.

Workflow 3: Research assistance

Refining a research question:

> I am trying to formulate a research question about the role of women
  in Cathar communities in 13th-century Languedoc.
  My current formulation: [paste].
  What is unclear or underspecified? What assumptions am I making?
  Suggest 3 alternative formulations at different levels of scope.

Literature gap analysis:

> I have reading notes in /literature_notes/
  Based on what these articles cover, what topics or approaches
  seem underrepresented in this literature?
  What questions do these authors collectively not ask?

Identifying connections:

> Read my notes in /notes/ and the article article.pdf
  Does this article connect to anything in my notes?
  Are there arguments here that support, contradict, or extend
  things I am already thinking about?

Preparing for supervision or a seminar:

> Read chapter_draft.docx
  Anticipate 5 critical questions a supervisor might raise about this chapter.
  For each question, suggest how I might respond.

Workflow 4: Extracting structured data

This is where Claude Code most clearly outperforms Desktop for researchers with any quantity of sources.

From a single document:

> Read deposition_guilhem.pdf
  Extract every person mentioned. For each person:

  - Name as it appears in the text

  - Their relation to the deponent (if stated)

  - What they are accused of or associated with
  Output as a table in persons_guilhem.md

Across a corpus:

> I have 30 deposition files in /transcriptions/ (all .docx)
  For each file, extract:

  - Deponent name

  - Date of deposition

  - Location

  - Names of persons mentioned
  Compile everything into depositions.csv

From a structured source:

> Read the_register.xlsx
  Find all entries where the "outcome" column is "absolution"
  and the "date" column is before 1280.
  List them with the person's name and the source reference.

With instructions for handling uncertainty:

> When you are not sure about a name, write it with [?] rather than guessing.
  If a date is given as a feast day, convert it and note the conversion.
  If text is missing or illegible, write [gap].

Workflow 5: Organising and renaming files

Claude Code can work with your filesystem — not just file contents but file names and folder structure.

Renaming files consistently:

> List all files in this folder.
  Rename them to follow the pattern: YYYY_author_shorttitle
  using the information in each file's first page.
  Show me the proposed new names before renaming anything.

Auditing a folder:

> List all files in /sources/
  Tell me: how many are PDFs, how many are DOCX, how many are other formats?
  Are there any duplicates (same name, different folder)?

Moving files into subfolders:

> I have a folder with 60 PDFs. They are all named with a year at the start
  (e.g. 1245_montauban.pdf, 1312_carcassonne.pdf).
  Create subfolders by decade (1240s, 1250s, etc.) and move each file
  into the appropriate subfolder.

Always ask Claude to show the plan before acting on files:

> Before you move or rename anything, list what you plan to do.
  Wait for my confirmation.

Workflow 6: Format conversion

Claude Code can convert between formats — no specialist software needed for many common cases.

Markdown to Word:

> Read my_notes.md and save it as my_notes.docx

Word to Markdown:

> Read draft.docx and save the text as draft.md
  (plain text, preserve headings and paragraphs, drop formatting)

CSV to readable table:

> Read data.csv and format it as a markdown table. Save to data_table.md

Multiple files to one:

> Read all .md files in /chapter_sections/ and combine them into
  one file called full_chapter.md, in alphabetical order by filename.

For more complex conversions (e.g., markdown to properly styled Word with a template), Claude can run Pandoc — a free conversion tool — if it is installed:

> Convert notes.md to notes.docx using Pandoc with the academic template.

Claude will run the command for you. You do not need to know the Pandoc syntax.

Workflow 7: Project organisation and tracking

Setting up a project structure:

> I am starting a new research project on Waldensian communities
  in 14th-century Italy.
  Create a folder structure appropriate for a document-heavy archival project.
  Create a CLAUDE.md template I can fill in.
  Create an empty progress.md.

Weekly task planning:

> Read progress.md and CLAUDE.md
  I have about 12 hours of research time this week.
  I need to finish the first draft of section 2 by Friday.
  What are the most important things to do this week?
  Draft a day-by-day plan.

Reviewing your own notes:

> Read everything in /notes/
  What are the main themes that keep coming up?
  Are there any contradictions between different notes?
  What questions do my notes raise that I have not yet addressed?

Summarising progress for a collaborator:

> Read progress.md and the files in /output/
  Write a 200-word update I can send to my project PI
  summarising what has been done and what the next steps are.

Workflow 8: Working with your Zotero library

If you have connected the Zotero MCP server (see A.setup.settings-local and A10.claude-and-zotero), Claude can search and use your library:

> Search my Zotero library for sources on social networks in medieval heresy.
  List the 5 most relevant items with author, title, and year.

> I am writing about the Cathar perfecti. What sources in my Zotero library
  are most relevant? Summarise how each one relates to the topic.

> Find sources in my Zotero library from the last 5 years on network analysis
  in historical research. Are there any I haven't read yet (no notes attached)?

No code. Zotero does not need to be open. Claude searches your library and reports back.

Workflow 9: Reading and processing web content

Claude Code can fetch and read web pages:

> Fetch the page at [URL] and summarise what it says about [topic].

> The call for papers for [conference] is at [URL].
  Read it and tell me: what are the key themes, the deadline, and
  whether my research on [topic] would be a good fit.

> Search the web for recent publications on Waldensian history since 2020.
  List titles, authors, and where they were published.

Note: Web search requires the tool to be enabled — either via a MCP server (Brave Search) or built-in web search capability. Check A.setup.settings-local if it is not working.

The pattern underlying all of these

Every workflow above follows the same structure:

  1. Where are the inputs? (a file, a folder, a URL, your Zotero library)
  2. What do you want done with them? (read, summarise, extract, convert, reorganise)
  3. What should the output look like? (a file, a CSV, a table, a draft, a list)
  4. Any constraints or conventions? (format, language, length, what to do when uncertain)

When your instruction covers all four, Claude can usually execute without needing to ask clarifying questions. When it fails or misunderstands, you can usually trace it back to one of these four things being ambiguous.

Under the hood: Claude writes and runs code for you

"Zero coding" means you write no code. It does not mean no code runs.

Claude Code is an agent: when you ask it to do something, it decides which tools to use. For many tasks — reading files, writing summaries, extracting text — it uses built-in tools that require nothing extra. But for more complex tasks, Claude will write a small script (usually Python), run it on your machine, read the result, and incorporate it into the response. You never see this code unless you ask.

Examples of when Claude writes code internally:

Task What Claude does behind the scenes
Summarise a PDF Uses its built-in Read tool — no code needed
Extract data from many PDFs into a CSV May write a Python script to loop through files
Analyse a spreadsheet (counts, filtering) May write a pandas script for reliability
Convert markdown to styled DOCX Runs Pandoc via shell command
Run OCR on a scanned PDF Runs OCRmyPDF or similar tool
Merge or split spreadsheets May write a Python + openpyxl script

This is a feature, not a complication. It means Claude can do things that would otherwise require you to learn programming — it writes the program, runs it, and gives you the result.

For one-off tasks, not seeing the code is fine — you asked for a result, not a program. For anything you might call an analysis — a visualisation, a data transformation, something that will end up described in a methods section — it is worth explicitly asking Claude to keep the script rather than discard it. The script is not scaffolding you throw away once you have the output; it is the reproducible record of what the analysis actually did, and it can be rerun, checked, or handed to a collaborator later. See B.lifecycle.4.dataanalysis for why this matters for the execution/interpretation distinction.

What this means practically: Some tasks work better or become available only when certain tools are installed on your machine. The most useful one is Python.

Installing Python for Claude Code use

Python is the language Claude most commonly reaches for when it needs to write a script. It is free, widely available, and runs on all platforms.

Do you already have Python?

Open a terminal and type:

python3 --version

If you see something like Python 3.11.4, you are set. Skip to the next section.

If you see "command not found" or a version below 3.8, install it.

Installing Python: the recommended approach — uv

The simplest way to install and manage Python for Claude Code use is uv — a modern Python manager that is fast, self-contained, and does not require you to understand Python packaging.

Mac / Linux — install uv:

curl -LsSf https://astral.sh/uv/install.sh | sh

Then close and reopen the terminal. Now install Python:

uv python install 3.12

That's it. Python 3.12 is now available. Claude Code will find it automatically.

Windows — install uv:

powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"

Then close and reopen PowerShell. Install Python:

uv python install 3.12

Why uv rather than installing Python directly?

The "official" way to install Python (from python.org) works, but can cause problems: multiple Python versions conflicting, packages installed in the wrong place, PATH issues. uv handles all of this cleanly. For researchers who just want Python to be available without managing it, uv is the right choice.

Verifying the installation

After installing uv and Python, confirm it works:

python3 --version

You should see Python 3.12.x (or similar).

Python packages Claude commonly uses

When Claude writes a Python script, it may need libraries beyond the standard ones. The main ones relevant to research tasks:

Package Used for
pandas Spreadsheet and CSV work, data analysis
openpyxl Reading and writing Excel (.xlsx) files
python-docx Reading and writing Word (.docx) files
pymupdf (fitz) Advanced PDF processing
requests Fetching web pages

Claude can install these itself when it needs them, if you allow it (via the Bash tool permission). You will see something like:

● Running: pip install pandas

This is Claude installing what it needs. You can let it proceed. The packages are small and the installation is fast.

If you prefer to have them installed in advance:

pip install pandas openpyxl python-docx pymupdf requests

Or with uv:

uv pip install pandas openpyxl python-docx pymupdf requests

What does require some technical help

A few things in Claude Code do need a brief technical moment — but not from you, necessarily. A colleague who is technical can set these up once:

After that, everything in this document is available without further technical knowledge.

Related