Most retail investors run their next acquisition through a chat reply. The decision lands as a generalist summary, not as analysis. Three systems connected to Claude close that gap: Perplexity's MCP for live market data, NotebookLM for cross-document reasoning across the deal pack, and the Financial Services plugin for the underwriting layer. Three installs. One investment memo, one risk register, one recommendation.
An investment committee runs every deal through the same three questions. What does the market say. What does the document set say. What do the numbers actually do under stress.
A retail investor running the same deal through a single Claude chat reply gets one answer to all three, blended into a paragraph. The market view, the document review, and the underwriting collapse into a summary. The summary reads convincing. The decision behind it is undefended.
The three systems below split the question back out. Perplexity's MCP gives Claude a live market view that is sourced, not memorised. NotebookLM lets Claude reason across the full deal pack rather than each document in isolation. The Financial Services plugin runs the cash-on-cash, yield, and stress-test layer that a generalist chat reply skips.
Three systems, one conversation, one decision. The output is closer to an investment memo than a chat reply.
An IC asks three questions. A chat reply gives one blended answer. Three systems split the answer back out.
One Claude conversation that runs your next off-plan deal through three systems in sequence and delivers three artefacts at the end. An investment memo, a risk register, and a recommendation with conditions.
Foundation: Claude with three connections wired in. Perplexity's MCP for live market data. NotebookLM linked as the document source for the deal pack. The Financial Services plugin enabled for the underwriting layer.
Inputs: one deal pack uploaded to NotebookLM. Brochure, payment plan, floorplan, master plan, and any market notes the broker has shared. Plus the deal terms in one prompt: price, payment plan, expected rent, holding period, exit assumption.
Output: an investment memo that names the comparables and the market view, a risk register that names the document-level risks, and a recommendation that names the conditions under which the deal works and the conditions under which it does not.
Perplexity's MCP installed in Claude. NotebookLM connected and loaded with the deal pack. The Financial Services plugin enabled. This is the only configuration step in the whole build.
The three connections stay live across conversations. The deal-by-deal work is the prompts, not the setup.
One prompt asks Claude to use the Perplexity MCP to pull live market data on the submarket: supply pipeline, recent transaction prices, comparable project activity, rental indices, and any visible regulatory or macro context.
Claude returns a sourced market view, not a memorised one. Every claim is anchored to a Perplexity citation.
One prompt asks Claude to reason across the full NotebookLM source set, not each document in isolation. The brochure's claims get checked against the payment plan. The floorplan gets checked against the unit mix in the master plan. The Perplexity research from layer two gets added to the same NotebookLM source set so Claude reasons across it too.
The output is a document-level risk register that names the discrepancies, the gaps, and the unverified claims.
One final prompt asks Claude to use the Financial Services plugin to underwrite the deal: cash-on-cash, gross and net yield, total cash to close, and a stress test against rent void, exit-cap-rate widening, and a delayed handover.
Claude combines the market view from layer two, the document review from layer three, and the underwriting from layer four into one investment memo with a recommendation and conditions.
Claude Pro, Max, Team, or Enterprise. MCP connections and plugins are available on these tiers as of December 2025. The free tier does not include MCP or plugin access.
A Perplexity account with API access. The Perplexity MCP authenticates against a Perplexity Pro or API account. Pro is the cheaper route for the prompt volume of one deal a week.
A Google account for NotebookLM. NotebookLM is free at notebooklm.google.com. Each deal gets its own notebook, scoped to that deal's documents.
Thirty minutes for the first build end to end. After that, each new deal is roughly fifteen minutes to a finished memo.
The Financial Services plugin is gated by region and account type. Check the plugin store in Claude before you start. If the plugin is unavailable in your region, the workflow still runs end to end. Claude can do the underwriting in plain Python inside the same conversation, but the output is less polished than the plugin's native templates.
Install the Perplexity MCP in Claude. Connect a NotebookLM notebook scoped to the deal pack. Enable the Financial Services plugin. After this, every step is a prompt in the same conversation.
One. Install the Perplexity MCP. In Claude, open Settings, go to Connectors, and add the Perplexity MCP. Authenticate with a Perplexity Pro account or an API key. Once installed, the connection persists across conversations.
Two. Create the deal notebook in NotebookLM. Open notebooklm.google.com and create a new notebook for this deal. Upload the brochure, payment plan, floorplan, master plan, and any market notes. Connect the notebook to Claude through the NotebookLM connector in Settings.
Three. Enable the Financial Services plugin. In Claude, open the plugin store, find Financial Services, and enable it for this workspace. The plugin gives Claude underwriting templates for cash-on-cash, yield, stress test, and sensitivity analysis.
MCP connectors and the plugin store are available on Claude Pro, Max, Team, and Enterprise. The free tier does not surface either menu. If you do not see Connectors in Settings, check your plan first before troubleshooting the install.
Each new deal gets a fresh NotebookLM notebook. Mixing documents across deals dilutes the source set and the cross-document reasoning. One notebook, one deal pack, one conversation. The setup work above runs once. The notebook work repeats per deal and takes about three minutes.
One message asks Claude to use the Perplexity MCP to pull the live market view on the submarket the deal sits in. Supply pipeline. Recent transaction prices. Comparable project activity. Rental indices. Any visible regulatory or macro context. The output is a market view anchored to citations, not memory.
Claude's training data has a cut-off date. A market view built from memory is always behind the live market. The Perplexity MCP closes the gap by routing the research through a live web index. The citations are the proof. Every claim in the market view should be traceable back to a URL, a publisher, and a date.
A market view built on price and yield alone reads convincing and underwrites badly. Supply pipeline and comparable project activity tell you whether today's price holds. Regulatory and macro context tell you whether the holding period assumption is realistic. Cutting dimensions to keep the response short makes the memo prettier and the recommendation weaker.
Once Claude has the market view, switch the source to the NotebookLM notebook holding the deal pack. The prompt asks Claude to reason across every document together, not each in isolation. The output is a risk register that names discrepancies, gaps, and unverified claims.
The default failure mode of an LLM with multiple documents is to summarise each one in turn. That is not reasoning, that is a table of contents. The phrase "reason across every document together" in the prompt is what flips the behaviour. Without it, you get a polite recap. With it, you get the discrepancy between page four of the brochure and clause six of the payment plan.
The second sentence of the prompt asks Claude to add the previous message's Perplexity research to the NotebookLM source set. This is the link between layer two and layer three. Without it, the document review happens in isolation from the market view and the risk register misses the comparable-project context.
The final prompt asks Claude to use the Financial Services plugin to underwrite the deal, then combine the market view from layer two, the risk register from layer three, and the underwriting into a single investment memo with a recommendation and conditions.
A memo that ends with "proceed" or "do not proceed" is useful once. A memo that ends with "proceed if the developer guarantees handover by Q4 and the rent void assumption holds at eight percent" is useful every time the deal changes. The conditions are the audit trail for the decision, and they are what the broker has to negotiate against.
If the plugin is not available in your region or on your plan, swap the first half of the prompt for: "Use a Python tool inside this conversation to underwrite the deal with the same line items below." Claude runs the maths in Python, returns the same numbers, and the memo composes the same way. The plugin polishes the output. It does not unlock anything Claude cannot do otherwise.
Claude Pro at twenty dollars a month. Perplexity Pro at twenty dollars a month, optional if you already have API access on another plan. NotebookLM is free. The Financial Services plugin is included in your Claude tier. No metered runtime, no per-deal billing.
Pro at twenty dollars a month covers MCP connectors and the plugin store. Step up to Max only if you are running multiple deals in parallel or hitting message limits during long memo passes.
Perplexity Pro at twenty dollars a month authenticates the MCP for normal-volume use. One sourced market view per deal sits well inside the allowance.
NotebookLM is free for personal use at notebooklm.google.com. One notebook per deal. The free quota supports the document volume of a typical off-plan pack with margin to spare.
The stack is good at structuring a decision. It is not the decision. The five things below are where the investor still has to do the work.
Perplexity returns citations, not truth. A broker blog and a research house report both show up as sources. Read the citation list before trusting the market view. Two or three of the top-cited URLs should come from publishers you would already trust on paper.
NotebookLM reads what is in the deal pack. It cannot read what the developer left out. The escrow problem on an earlier project, the snagging history a previous buyer logged, the planning condition lifted in a council meeting. These need a separate call to a lawyer and a search of the public record.
Cash-on-cash and IRR are only as good as the rent and exit-cap-rate assumptions. The plugin runs the maths fast. It does not validate the inputs. If your rent assumption is off by 15 percent, the memo's recommendation is off by 15 percent. Pressure-test the inputs against the market view from layer two before signing off.
An IC-grade memo is a tool for getting to a yes or a no with a partner, a co-investor, a spouse, or a broker. The reader still has to push back, ask the conditions, and walk the unit. A memo no one challenges is a memo that ships the wrong deal cleanly.
Three systems compress a process that used to take a week into a process that takes thirty minutes. That compression is also the risk. The discipline that used to sit in the calendar now has to sit in the prompts. Use the conditions section at the bottom of the memo to slow yourself down before pressing send.
Day 13 watches the portfolio you already own through one live artefact that refreshes itself every month. Day 14 sits in front of every new deal as a decision gate. One watches today's holdings. One underwrites tomorrow's acquisition.
A retail investor running Day 13 alone sees the portfolio change in real time. Add Day 14 in front of every new acquisition and every new deal goes through the same three-system review. Two conversations, one shared market view, one operator. The portfolio that gets watched and the deals that get added both live to the same standard.
By the end of thirty minutes, the chat has a sourced market view from Perplexity, a document-level risk register from NotebookLM, an underwriting table from the Financial Services plugin, and a synthesised memo with a recommendation and conditions. The memo names the comparables, names the risks, names the numbers, and names the conditions under which the recommendation flips.
The next deal does not start with you typing "is this a good deal" into Claude. It starts with you handing Claude the deal pack and reading what the IC stack already produced. That is the architecture shift Day 14 ships.
Every prompt in this 30-day series is one capability. FourthspaceOS bundles all of them into a single product: underwriting, comps, market research, deal sourcing, portfolio tracking, and investor reporting. The product runs natively on the same Anthropic agents and plugins you are learning to use this month.
Waitlist members get founding pricing, early access, and one new prompt delivered each day for the next 30 days.
Founding pricing locks in for waitlist members.
One email a day for 30 days. No spam. Unsubscribe whenever.
Send them this page. The work it takes to write a real investment memo by hand is the work the three-system stack removes. The leverage is not the memo. It is the fact that every deal goes through the same three questions before it gets a yes.
Find me on Instagram ↗You do not need a live deal to test the workflow. Take the last off-plan project you reviewed and rerun it through the three systems. Compare the memo Claude produces to the call you actually made. The gap between the two is the part of your underwriting that was implicit until now.
Jump to setup ↑