Why We're Building an MCP Server for Deal Sourcing

Every investor has a thesis. A set of beliefs about where value will be created, which teams can capture it, and what stage is the right entry point.

Every deal sourcing tool ignores it.

You get filters. Industry dropdowns. Score ranges. Maybe a saved search that emails you a CSV once a week. The same results everyone else sees, because the tool doesn't know you.

We think deal sourcing should work differently. And the technology to make it happen just arrived.

The Problem with Search-Based Deal Sourcing

Current tools treat deal sourcing like e-commerce: browse a catalog, apply filters, scroll through results. This works for buying shoes. It doesn't work for finding your next investment.

Here's why:

Filters are lossy. When you filter for "fintech, seed stage, US-based," you lose the AI infrastructure company in Berlin that's solving a payments problem better than any US fintech. Your thesis would love it — but your filters killed it.

Results aren't ranked by fit. A list sorted by score or date tells you nothing about alignment. The #1 scored startup might be a terrible fit for your fund. The #47 might be exactly what you're looking for.

Context is missing. Your portfolio already has three dev tools companies. Do you really want a fourth? Or do you need something in adjacent infrastructure to create cross-portfolio value? Current tools don't know, so they can't tell you.

Every query starts from zero. No memory. No learning. No understanding of why you passed on the last 50 startups. Every session is a blank slate.

What Changes with MCP

MCP — Model Context Protocol — is an open standard that lets AI assistants connect to external data sources. Think of it as USB for AI: a universal plug that lets Claude, GPT, or any compatible assistant talk to structured APIs.

When an investor connects Athanor's MCP server to their AI assistant, the interaction changes fundamentally:

Before (keyword search):

"Show me fintech startups with score above 70"

After (conversational, personalized):

"Find startups that align with my thesis, complement my portfolio, and match the patterns of my best investments"

The difference isn't cosmetic. The second query requires the system to know:

1. What your thesis actually says
2. What's already in your portfolio
3. Which of your investments performed well — and what they had in common
4. What you've looked at and passed on — and why

This is what we're building.

Investor DNA: The Personalization Engine

We call it Investor DNA — a profile built from three layers:

1. Thesis (What You Say You Want)

The explicit part. Your stated investment preferences:

• Sectors: AI infrastructure, developer tools, vertical SaaS
• Stages: Pre-seed to Seed
• Check size: $100K – $500K
• Geography: US, EU
• Team preferences: Technical founders, repeat founders
• Avoid: Hardware, crypto, consumer social

This is table stakes. Every CRM has a version of this. But it's only 40% of the picture.

2. Portfolio (What You've Already Done)

Your existing investments tell a story your thesis statement doesn't:

• Synergy potential: Which new startups could create value with your existing portfolio companies?
• Gap analysis: Where is your portfolio underweight?
• Overlap risk: Would this investment compete with something you already back?
• Pattern recognition: What do your best-performing investments have in common?

Your portfolio is your revealed preference. We weight it at 30%.

3. Behavior (What You Actually Do)

The implicit layer. Over time, your search patterns, time-on-page, comparisons, and pass decisions build a behavioral signal:

• Startups similar to ones you spent time on → boost
• Startups similar to ones you passed quickly → deprioritize
• Patterns in your successful investments → amplify

No one fills out a form describing their unconscious biases. But behavior captures them. The remaining 30%.

What This Looks Like in Practice

Here's a real conversation between an investor and their AI assistant with Athanor MCP connected:

Investor: "What came in this week that I should look at?"

AI: Queries Athanor MCP with investor's thesis + portfolio context

3 new startups match your profile this week:

1. Relay — API infrastructure for financial data aggregation. 87% thesis alignment. Would fill a gap in your portfolio (no data layer investments). Seed stage, Bay Area, raising $2M.

2. Archway — AI-powered contract review for SMBs. 74% alignment. Strong overlap with your LegalStack investment — consider synergy angle. Pre-seed, remote, raising $800K.

3. Canopy — Vertical SaaS for arborists. 71% alignment. Unusual vertical but matches your "unsexy vertical + strong unit economics" pattern from GreenOps and ParkBase.

Investor: "Tell me more about Relay. How does it compare to Plaid?"

AI: Calls get_startup + compare tools

Relay is building an aggregation layer specifically for real-time payment data... Key differentiator from Plaid: focuses on write access (initiating payments), not just read access (account data)... GemScore: Potential 78, Readiness 65. Risk flags: single technical founder, no revenue yet...

Investor: "Compare Relay and that other fintech I looked at last week — the one from London."

AI: Calls compare_startups with history context

Comparing Relay (Bay Area) vs. Clearway (London):

Dimension	Relay	Clearway
Thesis fit	87%	62%
Portfolio fit	High (fills gap)	Medium (adjacent)
Team	Solo technical	2 co-founders, banking background
...

No filters. No dropdowns. No CSV exports. Just a conversation that gets smarter every time.

Six Tools, One Protocol

The MCP server exposes six tools that any AI assistant can call:

Plus persistent resources: alert subscriptions, watchlists, your investor profile, and interaction history.

The tools are composable. Your AI assistant chains them naturally: search → compare → analyze fit → track deal. No workflow to learn. No UI to navigate. Just ask.

Smart Alerts: Deals Come to You

The best deal sourcing doesn't require you to search at all.

Save your criteria once. Get notified when a matching startup appears:

• Thesis match: High-alignment startup gets evaluated → immediate alert
• Score change: Watched startup's GemScore shifts → notification
• Portfolio signal: Startup complements your portfolio → flagged
• New evaluation: Startup in your sector completes full GemScore → digest

Delivered however you want: in your next AI conversation, via email, webhook to Slack, or daily digest.

The investor who gets a notification at 9 AM that a high-fit startup just completed evaluation — and can review, compare, and reach out before lunch — has a structural advantage over the one browsing a dashboard once a week.

Why MCP, Why Now

Three things converged:

1. MCP reached critical mass. Claude Desktop, Cursor, Windsurf, and a growing list of AI tools support MCP natively. Investors already using AI assistants can add Athanor as a data source without switching tools.

2. AI assistants got good enough. Natural language queries over structured data require strong reasoning. Today's models can translate "find startups like the ones I've backed, but in adjacent markets" into the right combination of API calls. A year ago, they couldn't reliably.

3. We have the data. Athanor has evaluated hundreds of startups with deep, structured analysis — team, market, business model, risk, scores. This isn't scraping Crunchbase. It's proprietary evaluation data with 50+ structured dimensions per startup.

The intersection: a mature protocol, capable models, and a rich data layer. Building an MCP server isn't a research project anymore — it's an engineering one.

What We're Not Building

Worth being clear about boundaries:

• Not a CRM. We won't replace Affinity or Attio. But we will push data into them.
• Not a database dump. Only public, opted-in startups with completed evaluations are discoverable. No drafts, no private submissions.
• Not surveillance. Your investor profile is encrypted. Startups don't see who searched for them. Your thesis stays yours.
• Not a replacement for judgment. AI narrows the funnel. You still pick the winners. The goal is fewer hours on filtering, more hours on founder conversations.

The Roadmap

We're building in four phases:

Phase 1 is already in development. The static page at /investors/ai-discovery has a waitlist for early access.

The Bigger Picture

The MCP server isn't just a deal sourcing tool. It's the foundation for everything that comes next:

• Abyss (our two-sided marketplace) will use the same personalization engine to match investors with private deal flow
• GemScore V4 will feed real-time, living evaluation data into discovery — not static snapshots
• Investor DNA Matching will let founders find investors whose thesis matches their startup, not just the other way around

We're building the infrastructure for AI-native investing. The MCP server is the first public interface.

Getting Early Access

If you're an investor who already uses Claude, GPT, or any AI assistant in your workflow:

1. Join the waitlist at /investors/ai-discovery
2. We'll onboard early adopters in batches as each phase ships
3. Early access investors help shape the product — your feedback directly influences what we build

The first version is free during beta. Production pricing starts at $9/month for individual investors, with fund-level plans for teams.

The best deals aren't found by searching harder. They're found by searching smarter — with tools that know what you're actually looking for.

— The Athanor Team