Learning with MCP

Once your AI assistant is connected via MCP, you can use it to actively learn from your sources. This page covers the main workflows: exploring sources, Feynman technique review, and Socratic dialogue.

Exploring your sources

The simplest starting point: ask your AI assistant what you've been reading.

Example prompts:

• "List my sources and give me an overview of what I've been reading lately."
• "Search my sources for anything related to attention mechanisms."
• "Read my latest paper and give me a one-paragraph summary."
• "Compare these two papers — what do they agree on, and where do they diverge?"

Select sources in the Studio tab sidebar before asking, then your AI uses get_app_context to read them. Or let it search with search_sources_semantic to find relevant sources by topic.

Feynman technique sessions

The Feynman technique is simple: explain a concept from scratch, as if teaching it. If you can't explain it simply, you don't understand it yet.

FeynmanLM's MCP tools support structured Feynman sessions that track your understanding over time.

Starting a session

Ask your AI assistant to quiz you:

• "Start a Feynman session on CRISPR-Cas9 mechanisms."
• "Test my understanding of transformer self-attention."
• "Quiz me on the paper I just read about partial reprogramming."

The AI calls start_feynman_session, which searches your source library for relevant content and returns it as context. The AI then asks you to explain the concept.

During the session

The AI asks questions, you explain, and the AI evaluates your understanding:

1. You explain the concept in your own words
2. The AI evaluates — checking for accuracy, depth, and whether you addressed mechanisms vs. just surface-level facts
3. Follow-up probes — the AI asks targeted questions about gaps or imprecise parts of your explanation
4. Repeat until the topic is thoroughly covered

Each exchange is recorded via submit_feynman_answer with a score from 0.0 to 1.0.

Ending a session

When you're done, the AI calls end_feynman_session with an overall score and summary. You can review past sessions later with get_feynman_sessions to track your understanding progression.

Example session

You: Quiz me on oxidative phosphorylation.

AI: [starts Feynman session, loads your biochemistry sources] Explain the chemiosmotic theory — how does the proton gradient drive ATP synthesis?

You: So the electron transport chain pumps protons across the inner mitochondrial membrane, creating a gradient. Then ATP synthase uses that gradient to generate ATP.

AI: Good start — you've got the high-level flow right. But let me probe deeper: why does the proton gradient specifically drive ATP synthase rotation? What's the physical mechanism that converts a concentration gradient into mechanical work?

You: The protons flow through a channel in ATP synthase, and the flow causes a rotor subunit to spin. Each rotation mechanically changes the conformation of the catalytic sites, forcing ADP and phosphate together.

AI: Excellent — you've nailed the rotary catalysis mechanism. Score: 0.9. One more: what happens to ATP production if you add an uncoupling agent like DNP?

Socratic dialogue

Beyond structured Feynman sessions, you can have free-form Socratic dialogue with your AI about any source:

• "I just read this paper on diffusion models. Challenge my understanding — ask me hard questions."
• "I think I understand how RLHF works, but test me. Don't go easy."
• "Play devil's advocate on the claims in this source about longevity interventions."

The AI has full access to your source text, so it can cite specific passages, catch misunderstandings, and push you to think more precisely.

Tips

• Select sources first. Before starting a session, select the relevant sources in Studio so the AI has the full text as context.
• Be specific about topics. "Quiz me on attention" is less effective than "Quiz me on multi-head attention scaling and why we divide by sqrt(d_k)."
• Don't just answer — explain. The Feynman technique works because it forces you to articulate mechanisms, not just recall facts.
• Review past sessions. Ask "Show me my past Feynman sessions on [topic]" to see how your understanding has progressed.
• Use across sources. The best learning happens when you connect ideas: "I've read papers on both CRISPR and base editing — quiz me on the differences in their mechanisms."