Skip to content
Read • Question • ConnectAnthony’s notes & researchReturn with a new question
Anthony’s NotesA personal field guide
From the notebook

Library Processing

You can set this up at three levels. For your use case — PDF library + Obsidian + reading plans — I would start with Level 1, then move to Level 2 once the workflow is useful.

The core architecture

What you want is not one magic agent. You want a pipeline:

PDF Library
→ searchable index / retrieval
→ AI analysis
→ claim extraction
→ contradiction detection
→ Obsidian note generation
→ reading plan
→ critique/review pass

The “agent” is the AI system coordinating those steps.

Level 1: Easiest setup using ChatGPT or Claude

This is the lowest-friction version.

Option A: ChatGPT setup

Use this if you want the simplest workflow inside ChatGPT.

Step 1: Put your PDFs in Google Drive

Create a clean folder:

Google Drive/
  AI Research Library/
    Christianity and Western Morality/
    Enlightenment and Human Rights/
    Foreign Policy/
    Philosophy/
    Fitness and Health/
    Obsidian and PKM/

Use clear filenames:

Holland - Dominion.pdf
Siedentop - Inventing the Individual.pdf
Freeman - The Closing of the Western Mind.pdf
Nietzsche - Genealogy of Morals.pdf
SEP - Transcendental Arguments.pdf

Step 2: Connect Google Drive to ChatGPT

Go to:

ChatGPT → Settings → Apps → Google Drive → Connect

OpenAI’s official help page for Google Drive app sync is here:

https://help.openai.com/en/articles/10948259-google-drive-app-with-sync-self-service-setup

One important limitation: OpenAI’s help page for ChatGPT Agent says Agent Mode cannot access data from apps with sync like Google Drive, though Google Drive can still be used in chat and Deep Research when enabled. Source: https://help.openai.com/en/articles/11752874-chatgpt-agent

So for your PDF library, ChatGPT Deep Research / connected-file chat may be better than Agent Mode if the files live in Google Drive.

Step 3: Run the workflow manually in stages

Ask ChatGPT:

Search my connected Google Drive PDF library for documents about Christianity, Western morality, human rights, Enlightenment morality, natural law, and Nietzsche.
 
First, identify the relevant files. Do not summarize yet.
 
Return:
1. File title
2. Author
3. Main topic
4. Why it is relevant
5. Whether it should be treated as primary, secondary, or background

Then ask:

Now group those documents by topic.
 
Use these categories:
1. Christianity as historical source of Western morality
2. Secular foundations of human rights
3. Natural law and medieval legal development
4. Enlightenment morality
5. Nietzschean critique of Christian morality
6. Greek/Roman antecedents
7. Modern liberalism
8. Critiques of the Christianity-thesis
 
For each group, list the strongest sources and explain why they belong there.

Then:

Now extract the major claims from each author.
 
For every claim, include:
1. Author
2. Source file
3. Claim
4. Evidence used
5. What the claim supports
6. What the claim challenges
7. Confidence level

Then:

Now compare the authors.
 
Find contradictions, tensions, and disagreements between them.
 
For each contradiction, include:
1. Author A’s claim
2. Author B’s claim
3. Why they conflict
4. Whether the conflict is historical, philosophical, theological, or methodological
5. Which side currently seems stronger based on the evidence
6. What I should read next to resolve it

Then:

Now create Obsidian-ready notes from this analysis.
 
For each note:
- Use Markdown
- Include YAML frontmatter
- Use [[internal links]]
- Include key claims
- Include counterclaims
- Include source references
- Include unresolved questions
- Include related notes
- Include tags

This is not fully autonomous, but it is reliable because you control each stage.

Option B: Claude setup

Claude is also good for this, especially with Projects.

Anthropic says you can add Google Drive files to a Claude Project by opening the project, going to the Files section, clicking +, selecting Drive, then adding files. Source:

https://support.claude.com/en/articles/10166901-use-google-workspace-connectors

Claude Projects can also use RAG when project knowledge gets large, expanding the amount of knowledge Claude can access. Source:

https://support.claude.com/en/articles/11473015-retrieval-augmented-generation-rag-for-projects

Claude project structure

Create a Claude Project called:

PDF Library Research Agent

Add project instructions:

You are my research-library agent.
 
Your job is to help organize my PDF and note library by topic, extract claims, compare authors, identify contradictions, generate Obsidian-ready notes, and create reading plans.
 
Always distinguish:
1. Historical influence
2. Philosophical justification
3. Theological claims
4. Institutional transmission
5. Political development
6. Modern moral validity
 
Do not merely summarize documents. Reason across documents.
 
When generating Obsidian notes:
- Use Markdown
- Use [[internal links]]
- Include tags
- Include source references
- Include unresolved questions
- Include related notes

Then add your PDFs or Google Drive documents into the Project files.

Level 2: Stronger no-code / low-code setup for Obsidian

This is what I would aim for if you want a serious long-term research system.

Google Drive = raw PDF archive
Obsidian = permanent notes and output
ChatGPT = deep research, synthesis, critique
Claude = project-based PDF reasoning
Local RAG = searchable private PDF index

The missing piece is local RAG.

RAG means Retrieval-Augmented Generation: instead of dumping your whole library into a model, a search system retrieves the relevant chunks first, then the model reasons over those chunks. The original RAG paper describes this as combining a language model with an external retriever/memory so answers can use explicit retrieved knowledge rather than relying only on model memory. Source:

https://arxiv.org/abs/2005.11401

Practical tools to consider

For local/private PDF search, use one of these:

Zotero + Better BibTeX + Obsidian
DevonThink
AnythingLLM
Khoj
Open WebUI + local documents
LlamaIndex
Dify
LangChain
Claude Desktop + MCP filesystem access

For your setup, the most practical non-coding path is probably:

Zotero for PDFs
+
Obsidian for notes
+
Claude/ChatGPT for reasoning
+
optional local RAG later

Level 3: Advanced local agent setup with Claude Desktop + MCP

This is the closest to a real “agent that works over my local files.”

MCP means Model Context Protocol. It lets Claude Desktop connect to tools such as local folders, databases, search indexes, or custom scripts. Anthropic describes local MCP servers as a way for Claude Desktop to access tools and resources from your local machine. Official Anthropic setup page:

https://support.claude.com/en/articles/10949351-getting-started-with-local-mcp-servers-on-claude-desktop

A filesystem MCP server can let Claude see selected local folders, such as your Obsidian vault or a PDF folder.

On your Mac, create:

/Users/Anthony/AI-Library/
  PDFs/
    Christianity-Western-Morality/
    Enlightenment-Human-Rights/
    Foreign-Policy/
    Philosophy/
  Extracted-Text/
  Claims/
  Obsidian-Output/
  Reading-Plans/

And your Obsidian vault:

/Users/Anthony/Obsidian/AnthonyVault/
  Sources/
  Literature Notes/
  Permanent Notes/
  MOCs/
  Reading Plans/

What the MCP version can do

With local file access, Claude can potentially:

Read Markdown files
Create new Obsidian notes
Edit existing notes
Search folders
Organize files
Generate reading plans
Create topic maps
Create claim indexes

But there are security tradeoffs. A 2025 security paper on MCP warns that the MCP ecosystem can expose users to risks such as malicious servers, tool hijacking, and sensitive data exfiltration if servers are not trusted. Source:

https://arxiv.org/abs/2510.16558

So only use MCP servers you trust, and restrict them to specific folders, not your whole Mac.

The actual “agent” workflow to run

Once your files are accessible to ChatGPT, Claude, or local RAG, run this sequence.

Stage 1: Inventory agent

Goal: identify what is in the library.

Prompt:

Act as my library inventory agent.
 
Search the available PDF library and create an inventory.
 
For each document, return:
1. Title
2. Author
3. Topic
4. Historical period
5. Main thesis
6. Relevant tags
7. Whether this belongs in my active research queue, reference library, or archive
 
Do not summarize deeply yet. First build the map of the library.

Stage 2: Topic-clustering agent

Goal: group documents by theme.

Prompt:

Act as my topic-clustering agent.
 
Using the library inventory, group the documents into major research clusters.
 
For each cluster, include:
1. Cluster name
2. Main question
3. Included authors/documents
4. Why these documents belong together
5. Internal subtopics
6. Which cluster should be studied first

For your current work, likely clusters would be:

Christianity and Western morality
Natural law and human rights
Greek/Roman antecedents
Medieval individualism
Reformation and conscience
Enlightenment liberalism
Nietzsche and slave morality
Secular moral realism
Foreign policy realism/restraint
AI and knowledge systems

Stage 3: Claim-extraction agent

Goal: turn books/articles into argument units.

Prompt:

Act as my claim-extraction agent.
 
For the selected topic cluster, extract the major claims from each document.
 
For each claim, include:
1. Claim
2. Author
3. Source document
4. Evidence used
5. Concepts involved
6. Opposing claim, if present
7. Related Obsidian note title
8. Tags

This creates the raw material for your Obsidian notes.

Stage 4: Contradiction-detection agent

Goal: compare authors.

Prompt:

Act as my contradiction-detection agent.
 
Compare the extracted claims across authors.
 
Find:
1. Direct contradictions
2. Methodological disagreements
3. Different uses of the same term
4. Historical disagreements
5. Philosophical disagreements
6. Places where authors talk past each other
7. Places where my own notes may be conflating separate issues
 
For each tension, explain:
1. Author A’s position
2. Author B’s position
3. The nature of the disagreement
4. What evidence would resolve it
5. Which additional source I should read

Stage 5: Obsidian note agent

Goal: create permanent notes.

Prompt:

Act as my Obsidian note agent.
 
Turn the claim analysis into Obsidian-ready notes.
 
Create:
1. One MOC note for the whole topic
2. One note per major author
3. One note per major concept
4. One note per major contradiction
5. One reading-plan note
 
Formatting rules:
- Use Markdown
- Use YAML frontmatter
- Use [[internal links]]
- Use tags
- Include source references
- Include questions for further research
- Include "Related Notes" at the end

Stage 6: Critic sub-agent

Goal: prevent hallucinated coherence.

Prompt:

Now act as a skeptical critic reviewing the previous output.
 
Find:
1. Unsupported claims
2. Overgeneralizations
3. False equivalences
4. Missing counterarguments
5. Places where the notes confuse historical origin with philosophical justification
6. Places where author positions may have been oversimplified
7. Missing primary sources
 
Return a corrected version of the analysis.

This is the “ask another sub-agent to critique the result” part. You do not need a separate AI account. You can simply make the same model switch roles.

Best first implementation for you

Do this first:

1. Pick one topic only:
   Christianity and Western Morality.
 
2. Create one Google Drive folder:
   AI Research Library / Christianity and Western Morality
 
3. Put 10–20 files in it:
   Holland, Siedentop, Freeman, Nietzsche, Himmelfarb, Taylor, Moyn, Tierney, SEP articles, your notes.
 
4. Connect the folder to ChatGPT or Claude.
 
5. Run the six-stage prompt sequence above.
 
6. Export the final notes into Obsidian.

Do not begin with your whole PDF library. Start with one research cluster. Once the workflow works, scale it.

The Obsidian output template

Use this for each generated note:

---
type: permanent-note
topic: Christianity and Western Morality
status: draft
created: 2026-06-15
tags:
  - morality
  - christianity
  - enlightenment
  - human-rights
sources:
  - 
---
 
# Note Title
 
## Core Claim
 
## Explanation
 
## Supporting Evidence
 
## Counterarguments
 
## Related Authors
 
## Contradictions / Tensions
 
## My Current View
 
## Open Questions
 
## Related Notes
 
- [[Christianity and Western Morality MOC]]
- [[Historical Influence vs Philosophical Justification]]
- [[Natural Law]]
- [[Human Rights]]

The reading-plan prompt

After the system has clustered your library, ask:

Create a reading plan from this library.
 
Goal:
Help me understand whether Christianity is the foundation of Western morality, and what the strongest secular counterarguments are.
 
Constraints:
- Prioritize high-signal sources
- Separate primary sources from secondary sources
- Put books/articles in the best order
- Include why each source comes before the next
- Include what question I should bring to each source
- Include what Obsidian note I should create after each reading
- Create a 4-week plan and a 12-week plan

My recommendation

Use this order:

First: Claude Project or ChatGPT connected Google Drive
Second: Obsidian output workflow
Third: Zotero for citation/library hygiene
Fourth: local RAG or MCP only after the first workflow proves useful

The biggest mistake would be trying to build a complex local agent system before you know exactly what outputs you want. Start with one topic cluster, make the process work, then automate.