Video: https://youtu.be/LIVuhDH6nM8 Repo: https://github.com/mspremulli/hackathon-yconic-team-ackm
Inspiration
The sponsor judges' opening talks highlighted a common goal: to create intelligent agents that can make meaningful, enterprise-grade decisions using real-world data. Combined with our research into each sponsor’s technology, we saw an opportunity to demonstrate how MCPs could revolutionize startup evaluation through autonomous AI workflows.
What it does
yc0n1c’s MCP Engines are autonomous AI VC systems that evaluate startups, score traction, analyze market conditions, and recommend funding—securely, scalably, and with explainable, real-time decisions. By integrating multiple sponsor tools into modular MCP servers, yc0n1c simulates what a fair, data-driven venture capitalist would do at hyperscale.
How yc0n1c Uses MCP Infrastructure to Power AI-Driven Startup Evaluation and Funding
To make startup evaluation transparent, scalable, and meritocratic, we built and integrated a suite of purpose-driven MCP (Modular Capability Protocol) servers and agents into yc0n1c, our AI venture capitalist assistant. These components work in harmony to enable yc0n1c to analyze, accept or reject startups, assess traction, determine funding amounts, and make autonomous, explainable investment decisions.
1. Market_Analyst MCP Server + Orchestrator Agent
We created and deployed the Market_Analyst MCP Server, which connects directly to our AWS Bedrock Agent core. An orchestrator agent intelligently calls this server when market insights are needed.
Claude and Mistral LLMs are used to synthesize structured evaluations of:
- TAM/SAM/SOM calculations
- Competitor analysis
- Market timing and trend velocity
- Risk exposure and market saturation
Tavily enhances this by performing real-time web-based startup research.
Senso.ai summarizes this data into clear, digestible recommendations for the yc0n1c decision engine.
Together, this stack allows yc0n1c to score how well-positioned a startup is within its market and identify external factors that should affect its funding potential.
2. Social Media MCP Server
We built a second MCP module focused on capturing external sentiment and social traction: the Social Media MCP Server.
It uses Bright Data to scrape and aggregate public social media data (bypassing the need for individual platform APIs).
The collected data is:
- Summarized by Claude or Mistral
- Analyzed for sentiment and traction signals (positive, neutral, negative)
- Stored in MongoDB for persistent evaluation context
This allows yc0n1c to assess whether a startup is receiving real-world attention, buzz, or customer validation—vital inputs to Proof of Scale analysis.
3. Core Infrastructure and Security
All MCPs interact through a secure and resilient architecture:
- MongoDB acts as the central datastore for parsed application data, traction metrics, and MCP outputs.
- Startups connect their Proof of Scale data through a client-facing MongoDB interface, which lets yc0n1c validate real milestone achievements and traction events.
- Temporal is used as a wrapper for all sensitive data flows (especially social media inputs), ensuring:
- Encryption
- Traceability
- Retry logic and resilience across the system
- Encryption
System Summary
| Component | Purpose |
|---|---|
| Market_Analyst MCP | Market position scoring, TAM/SAM/SOM estimation |
| Orchestrator Agent | Dynamic routing to Market MCP or Social MCP |
| Social Media MCP | Sentiment analysis, public traction signals |
| MongoDB | Central intelligence repository for startup data |
| Temporal | Secure data flow and transaction orchestration |
| Claude + Mistral | Multi-perspective reasoning and insight synthesis |
| Tavily + Bright Data | Market research + real-world data collection |
| Senso.ai | Intelligent data summarization for decisioning |
By combining these components into a modular, MCP-based architecture, yc0n1c is no longer just an assistant—it’s an intelligent, secure, and unbiased AI VC partner, capable of making funding decisions at scale, in real time, with explainable reasoning behind every move.
How we built it
We built two main MCP servers: the Market_Analyst MCP and the Social Media MCP, both integrated into our AWS Bedrock-based orchestrator agent. We used Claude, Mistral, Tavily, Bright Data, MongoDB, Temporal, and Senso.ai to extract, analyze, summarize, and securely route startup data for scoring and funding decisions.
Challenges we ran into
Integrating multiple third-party APIs and LLMs into a real-time evaluation loop required careful coordination and fallback logic. Building secure, resilient workflows with Temporal while maintaining explainable reasoning across agents pushed us to design beyond traditional hackathon MVPs.
Accomplishments that we're proud of
We created an autonomous VC agent that not only evaluates startups but also explains its decisions using multi-perspective LLM synthesis. We’re proud to have built an MCP system that simulates real-world investor logic using live, dynamic data feeds from multiple sponsor technologies.
What we learned
We learned how powerful MCP architecture becomes when applied to real-world use cases like funding decisions—especially when paired with secure workflows and intelligent orchestration. We also discovered that simplifying agent responsibilities and using MCPs as functional blocks enables scale, explainability, and modular growth.
What's next for yc0n1c’s AI-Driven Startup Evaluation & Funding MCP's
We plan to deploy yc0n1c with early-stage startup cohorts to gather live evaluation feedback and refine funding thresholds. Our roadmap includes tokenized milestone funding, full smart contract integration, and exposing MCP endpoints so founders can connect their Proof-of-Scale data directly to our agentic capital engine.
Log in or sign up for Devpost to join the conversation.