DealPilot — AI Sales Proposal Agent

Google Cloud Rapid Agent Hackathon · MongoDB Partner Track

DealPilot is an autonomous AI sales agent that turns discovery call transcripts into winning business proposals. It analyzes client needs, searches similar past projects in MongoDB Atlas, generates a tailored proposal with Gemini, and saves results for future reuse.

Built with Google ADK (Agent Builder), Gemini, and MongoDB MCP integration.

---

Demo flow

1. Paste a discovery call transcript (or click Load sample).
2. The agent runs a transparent 3-step pipeline:
   • Step 1 — Analyzer: Extract industry, company size, pain points, timeline, budget.
   • Step 2 — Researcher: Query MongoDB for similar historical projects.
   • Step 3 — Generator: Write an 800–1200 word professional proposal.
3. View results as plain text or export a PDF via the Anvil API.
4. Proposals are auto-saved to MongoDB Atlas.

---

Architecture

flowchart LR
    UI[Next.js Frontend :3000]
    ADK[Google ADK Agent :8080]
    MCP[MongoDB MCP Server]
    API[FastAPI Backend :8000]
    DB[(MongoDB Atlas)]

    UI -->|chat / run| ADK
    UI -->|save proposal| API
    ADK --> MCP
    MCP --> DB
    API --> DB
    UI -->|PDF export| Anvil[Anvil PDF API]

Loading

Layer	Technology	Port
Frontend	Next.js 16, React 19, Tailwind CSS 4	3000
Agent	Google ADK (dealpilot) + Gemini	8080
MCP	FastMCP server → MongoDB Atlas	stdio
Backend	FastAPI (save API + fallback pipeline)	8000
Database	MongoDB Atlas proposaldb	cloud

---

Features

• Multi-step agentic pipeline — plans and executes analyze → research → generate under user oversight
• MongoDB MCP tools — search projects, save proposals, list recent proposals, aggregate stats
• Chat UI — ADK-style interface with tool-call visibility and quick prompts
• Auto-save — generated proposals persist to the proposals collection
• Dual output — plain-text proposal or PDF download (Anvil)
• Fallback mode — FastAPI 3-step Gemini pipeline when ADK is unavailable

---

Prerequisites

• Python 3.11+
• Node.js 20+
• MongoDB Atlas cluster (free tier works)
• Google AI Studio API key (Gemini)
• Anvil API key (optional — PDF export; dev key is free with watermark)

---

Quick start

1. Clone and configure

git clone https://github.com/AaronTide/business-proposal-agent.git
cd business-proposal-agent

Copy environment files and add your keys:

# Backend
copy backend\.env.example backend\.env

# Agent
copy agent\dealpilot\.env.example agent\dealpilot\.env

# Frontend
copy frontend\.env.local.example frontend\.env.local

2. Backend setup

cd backend
python -m venv .venv
.\.venv\Scripts\pip install -r requirements.txt
.\.venv\Scripts\python seed_db.py

3. Agent setup

cd agent
.\scripts\start_agent.ps1

Opens the ADK web UI at http://127.0.0.1:8080 — select dealpilot from the dropdown.

4. Start backend API (MongoDB save)

In a new terminal:

cd backend
.\.venv\Scripts\uvicorn main:app --reload --port 8000

5. Start frontend

In a new terminal:

cd frontend
npm install
npm run dev

Open http://localhost:3000

---

Environment variables

backend/.env

Variable	Description
GEMINI_API_KEY	Google AI Studio API key
GEMINI_MODEL	Default: gemini-2.5-flash
MONGODB_URI	MongoDB Atlas connection string
MONGODB_DATABASE	Default: proposaldb

agent/dealpilot/.env

Variable	Description
GOOGLE_API_KEY	Gemini API key for ADK
GOOGLE_GENAI_USE_VERTEXAI	FALSE for AI Studio, TRUE for Vertex AI
MDB_MCP_CONNECTION_STRING	MongoDB Atlas URI (official MCP variable name)
MONGODB_DATABASE	Default: proposaldb
GEMINI_MODEL	Default: gemini-2.5-flash

frontend/.env.local

Variable	Description
ADK_AGENT_URL	ADK agent URL (default http://127.0.0.1:8080)
ADK_APP_NAME	Agent name: dealpilot
USE_ADK_AGENT	true = ADK agent, false = FastAPI fallback
BACKEND_URL	FastAPI URL for MongoDB save (default http://127.0.0.1:8000)
ANVIL_API_KEY	Anvil PDF generation API key

---

MongoDB MCP tools

The agent connects to MongoDB via a custom MCP server at backend/mcp_server.py:

Tool	Purpose
search_similar_projects	Find past projects by industry
find_documents	Query any collection with a filter
save_proposal / insert_document	Save generated proposal to proposals
list_recent_proposals	List latest saved proposals
aggregate_documents	Project stats by industry

Database collections

projects — seeded case studies (industry, description, outcome, budget)

proposals — saved generated proposals with requirements and timestamps

Seed the database

cd backend
.\.venv\Scripts\python seed_db.py

Test MCP tools locally

cd backend
.\.venv\Scripts\python scripts\test_mcp.py

---

Agent test commands

Try these in the chat UI or ADK web UI:

Find logistics projects in the database

Show me the 3 most recent saved proposals

Save this proposal to MongoDB

Or paste a discovery call transcript and let the agent run the full pipeline.

---

API endpoints

FastAPI (backend/main.py)

Method	Path	Description
POST	/proposal	Run 3-step Gemini pipeline (fallback)
POST	/api/v1/proposal/generate	Same pipeline, versioned
POST	/api/v1/proposal/save	Save proposal to MongoDB

Next.js API routes

Path	Description
/api/agent/chat	ADK session + auto-save
/api/proposal	ADK or FastAPI proposal generation
/api/anvil/pdf	Generate PDF from proposal text

---

Project structure

business-proposal-agent/
├── agent/
│   ├── dealpilot/
│   │   ├── agent.py          # Google ADK agent + MongoDB MCP
│   │   └── .env.example
│   ├── scripts/
│   │   └── start_agent.ps1
│   ├── system_prompt.md
│   └── mcp_setup.md
├── backend/
│   ├── main.py               # FastAPI API + fallback pipeline
│   ├── mcp_server.py         # MongoDB MCP server (FastMCP)
│   ├── seed_db.py            # Seed sample projects
│   └── scripts/test_mcp.py
├── frontend/
│   ├── app/
│   │   ├── api/agent/chat/   # ADK chat proxy
│   │   ├── api/anvil/pdf/    # PDF generation
│   │   └── page.tsx
│   └── components/
│       ├── AgentChat.tsx
│       └── ProposalOutput.tsx
└── README.md

---

Hackathon compliance

Requirement	Status
Functional agent beyond chat	✅ MongoDB tools + multi-step pipeline
Multi-step mission	✅ Analyze → Research → Generate
MongoDB MCP integration	✅ Custom MCP server on Atlas
Google ADK / Agent Builder	✅ ADK agent (dealpilot)
Gemini powered	✅ gemini-2.5-flash
Open-source license	✅ Add LICENSE file
Public hosted URL	⬜ Deploy to Vercel + Cloud Run
Demo video (~3 min)	⬜ Record and upload
Devpost submission	⬜ Complete form

Track: MongoDB Partner Track — Google Cloud Rapid Agent Hackathon

---

Troubleshooting

Agent not connecting to MongoDB

• Verify MONGODB_URI in backend/.env and agent/dealpilot/.env
• Run seed_db.py and scripts/test_mcp.py
• Restart the ADK agent after .env changes

Gemini 429 quota errors

• Free tier has daily limits; wait or use a new API key
• Enable the Gemini API in your Google Cloud project

PDF generation fails

• Set ANVIL_API_KEY in frontend/.env.local
• Restart the Next.js dev server after env changes

Official mongodb-mcp-server (npx) fails

• This project uses a custom FastMCP wrapper (backend/mcp_server.py) as a workaround
• See agent/mcp_setup.md for Agent Builder MCP configuration

---

Team

Built for the Google Cloud Rapid Agent Hackathon (MongoDB Partner Track).

Repository: github.com/AaronTide/business-proposal-agent