I Built a VC Brain for AI Agents (And You Can Install It in 30 Seconds)

Last week I was using Claude to draft a cold email to an investor. I found myself switching between three tabs — Claude for the email, NUVC for the investor’s thesis, and LinkedIn to figure out if they even do seed deals in Australia.

That felt broken. I built the investor database. Why can’t Claude just use it?

So I built an MCP server over the weekend. Now when I ask Claude "find investors who back AI startups at seed stage in Australia", it calls NUVC’s actual matching engine — 3,700+ investors, pre-computed embeddings, thesis alignment — and gives me real matches. Not hallucinated names. Real firms with real theses.

Today I’m releasing it publicly.

The Problem I Was Trying to Solve

If you’re a founder, you’ve probably asked ChatGPT or Claude something like "who invests in early-stage fintech in Southeast Asia?" and gotten a list that’s half-wrong, half-outdated, and entirely unhelpful. The AI doesn’t know which funds are actively deploying, what their actual thesis says, or whether they’d even look at your stage.

NUVC has this data. We’ve been building this investor database for over a year — 3,700+ investors, each with thesis, stage preferences, check sizes, geographic focus, track record. Every one of them has a 384-dimensional embedding so we can do semantic matching, not just keyword filtering.

The missing piece was getting that data into the tools founders actually use every day. And increasingly, that’s AI agents.

What Is MCP (And Why Should You Care)?

MCP (Model Context Protocol) is basically a universal plugin system for AI agents. Instead of every AI tool building its own integration layer, MCP gives them a shared protocol. Build one MCP server, and it works with Claude Desktop, Claude Code, Cursor, Windsurf, Cline — anything that speaks MCP.

For NUVC, this means one package gives every AI agent access to our full scoring engine, investor matching, and fund economics. One command to install. No API boilerplate.

What You Get: 8 Tools

I wanted these to cover the two questions every founder asks and the two questions every VC asks:

Founders ask: "Is my idea any good?" and "Who should I pitch?"

• nuvc_score — Scores your idea 0–10 across 5 VC dimensions. Same engine behind 250+ real investment memos.
• nuvc_match_investors — Finds best-fit investors from 3,700+ in our database. This is the one I wish I had when I was fundraising.
• nuvc_analyze — Market sizing, competitive landscape, full pitch evaluation.
• nuvc_roast — The brutally honest feedback your friends won’t give you.

VCs ask: "Is this deal worth a meeting?" and "Do these numbers make sense?"

• nuvc_score — Screen inbound deal flow. Score any pitch content in seconds.
• nuvc_venture_math — Burn multiple, LTV/CAC, Rule of 40, benchmarked against stage-appropriate thresholds.
• nuvc_fund_economics — Fee drag, GP alignment, J-curve, portfolio construction. The LP math.
• nuvc_extract — Pull structured data from any pitch deck, CIM, or deal memo.

How to Install (Genuinely 30 Seconds)

If you have Claude Desktop or Cursor:

npx @smithery/cli install nuvc --client claude

If you use Claude Code (what I use daily):

claude mcp add nuvc -- npx nuvc-mcp

Or just run it directly:

export NUVC_API_KEY=nuvc_your_key_here
npx nuvc-mcp

Free API key at nuvc.ai/api-platform/keys — 25 calls/month, no credit card.

See It In Action

Here’s what it actually looks like when you use it. These are real outputs from the NUVC engine, not mockups.

Scoring a startup idea

Me, in Claude:

Score this startup: We’re building an AI compliance platform that
auto-generates regulatory filings for fintech companies. $400K ARR,
3 enterprise customers, 2 founders with compliance backgrounds.

Claude calls nuvc_score and returns:

┌───────────────────────────────────────────────┐
│  NUVC VCGrade Score                            │
│                                               │
│  🟢 Overall: 7.4 / 10 — Strong                │
│                                               │
│  Problem & Market       8.1/10                │
│  Solution & Product     7.2/10                │
│  Business Model         7.5/10                │
│  Traction & Metrics     6.8/10                │
│  Team & Execution       7.9/10                │
│                                               │
│  Summary: Strong regulatory domain expertise  │
│  with early enterprise traction. ARR is light  │
│  for Series A but solid for seed extension.   │
└───────────────────────────────────────────────┘

That’s not Claude guessing. That’s the same engine that’s scored 250+ pitch decks for real accelerators and VCs.

Finding matching investors

Me:

Find seed-stage investors who back regtech/compliance startups in Australia

Claude calls nuvc_match_investors and returns:

## NUVC Investor Matching

5 investors matched from 3,700+ database | AI semantic matching

### 1. Sarah Chen — Apex Ventures
🟣 84% match (excellent)
Tier: Premier | Role: Leads Rounds | Check: $500K–$2M
Focus: RegTech, FinTech, Compliance
Thesis: We invest in regulatory infrastructure that
becomes the default for an industry...
Value-add: regulatory network, enterprise intros
AI Match: 82% | Fit boost: +2%

### 2. David Park — Southern Cross Capital
🟢 76% match (strong)
Tier: Established | Role: Lead or Follow | Check: $250K–$1M
Focus: FinTech, LegalTech, GovTech
Stages: seed, series_a
AI Match: 74% | Proven responder

[... 3 more matches]

───
Get full contact details and introductions at nuvc.ai

Those are real investor profiles from our database. Real theses. Real check sizes. Not hallucinated.

Quick venture math check

Me:

I have $400K ARR, $65K monthly burn, 14 months runway,
$180K CAC, $540K LTV. How do I look for Seed?

Claude calls nuvc_venture_math:

## NUVC Venture Math

| Metric               | Value  |
|----------------------|--------|
| Burn Multiple        | 1.95   |
| LTV/CAC Ratio        | 3.00   |
| Rule of 40           | 53.2   |
| Default Alive        | true   |
| Runway (months)      | 14     |
| Financial Health     | 7.8    |

### Benchmark vs SEED stage

| Metric          | Value | Signal      |
|-----------------|-------|-------------|
| Burn Multiple   | 1.95  | 🟢 Strong   |
| LTV/CAC Ratio   | 3.00  | 🟢 Strong   |
| Rule of 40      | 53.2  | 🟣 Elite    |
| Runway          | 14    | 🟡 Adequate |

The benchmarks are the real value here — NUVC compares your numbers against stage-appropriate thresholds. Claude could calculate burn multiple itself, but it can’t tell you whether 1.95x is good for Seed.

The Feature I’m Most Excited About

Honestly? It’s the investor matching.

Scoring is useful, but Claude is decent at giving general feedback on startup ideas. What it absolutely cannot do is search a real investor database. When you ask Claude to find investors, it makes them up. When you ask NUVC’s MCP, it queries 3,700+ real investor profiles using semantic embeddings that match on thesis alignment, not just keywords.

You get back: investor name, firm, match score, thesis snippet, check size range, whether they lead rounds, and their value-add tags. You don’t get contact details through the MCP (those are on the platform) — but you get enough to know who’s worth researching further.

I spent months building this matching engine for the NUVC platform. Having it available inside Claude feels like it finally found its natural home.

What’s Coming Next

This is v1.2.0. Three things I want to build next:

• Real percentile benchmarks — not generic blog-post thresholds. "Your $1.2M ARR puts you in the 72nd percentile of the 47 Series A decks we’ve scored." That requires our actual data, not math Claude can do locally.
• Deal Lens — same startup, different score depending on the investor type. An angel and a growth equity fund evaluate deals completely differently.
• Fund screening — for LPs evaluating VC funds against our 4,262-record fund library.

The guiding principle: if an AI agent can compute it locally, we shouldn’t wrap it in an API call. The MCP tools should give you access to NUVC’s proprietary data and intelligence — not be a calculator with a logo on it.

---

Try it: Get a free API key → npx nuvc-mcp → Ask Claude to score your startup idea.

If you build something cool with it, I’d genuinely love to see it. Find me on LinkedIn or email hello@nuvc.ai.

Links: npm · GitHub · Smithery · API Docs