AgentBridge-MCP

AgentBridge-MCP

MCP server for an AI assistant

AgentBridge-MCP

A modular, extensible Model Context Protocol (MCP) server for secure and auditable AI agent task routing.

---

🔧 What is This?

This repo is a foundational local MCP server built with Python + FastAPI that:

• Accepts structured JSON requests for tasks
• Routes them to secure modular task handlers (e.g. email, CSV parsing, summarization)
• Can be extended for any AI agent or tool

---

🚀 What You Get Out of the Box

✅ FastAPI MCP server
✅ Modular task/ folder (e.g. echo, get_time)
✅ Simple POST /mcp handler
✅ Dispatch logic with task map
✅ Ready for local testing & extension

---

💡 Why Use It?

This MCP server acts as the command center for AI agents.

It separates "what needs to be done" from "how it’s done" — enabling:

• 📦 Task modularity
• 🔐 A2SPA-compliant secure execution
• ⚡ Agent-Agnostic Logic (GPT, Claude, Ollama, etc.)

---

📁 Folder Structure

app/
├── main.py # FastAPI entry point
├── router.py # Dispatch logic
└── tasks/
├── echo.py
├── get_time.py
└── summarize_text.py

tests/
├── test_mcp.py

run.sh # Dev runner
requirements.txt # Dependencies

---

📊 Workflow Diagram

       ┌───────────────┐
       │  HTTP Client  │  (e.g. curl, frontend)
       └──────┬────────┘
              │  POST /mcp
       ┌──────▼───────┐
       │   main.py    │  (FastAPI app)
       └──────┬───────┘
              │
       ┌──────▼───────┐
       │  router.py   │  → Parses task key
       └──────┬───────┘
              │
       ┌──────▼────────────────────────┐
       │ app/tasks/{task}.py           │  → Executes logic (e.g., datetime, echo, LLM)
       └────────────┬──────────────────┘
                    │
             ┌──────▼───────┐
             │ Response JSON│
             └──────────────┘

---

📈 Coming Soon

This core server will soon power a Modular Investor Outreach AI Agent, with:

CSV-based email flows

Validation + personalization via MCP

PKI-audited A2SPA security

GPT/Ollama integration

---

🧠 Inspired By

Anthropic’s Claude MCP architecture

A2SPA Protocol (Agent-to-Agent Secure Protocol Architecture)

---

🛠 How to Run the Email Agent System

🔧 1. Install dependencies

pip install -r requirements.txt

📡 2. Start the MCP server (backend task router)

uvicorn app.main:app --reload

🧠 3. Optional: Start Ollama if not running already

ollama run llama3

🖥️ 4. Launch the Streamlit UI

streamlit run streamlit_app.py

💡 What This System Does

Upload a .csv of names and emails (columns: name,email)

Set a subject and shared body for the outreach message

Attach any files (pitch decks, PDFs, etc.)

Ollama personalizes each email with greeting and sign-off

MCP server dispatches email sending via Gmail SMTP
---

📁 File Requirements

Make sure .env is configured with:
EMAIL_USER=your@gmail.com
EMAIL_PASS=your_app_password # from Gmail App Passwords

---