AI Research Assistant using LangChain, LangGraph, LangSmith, RAG and MCP (Model Context Protocol)

Project Description

A research assistant that breaks a topic into subtopics, assigns research to agents, summarizes findings, and compiles a report.

Related MCP server: DeepResearch MCP

Features

• Graph-based agent orchestration with LangGraph

• Reproducible tracing with LangSmith

• Modular agent design for research tasks

  • Planner Agent: Breaks the topic into subtopics.

  • Researcher Agent: Gathers info for each subtopic.

  • Summarizer Agent: Summarizes and organizes into a report.

• Cache agent responses using SQLite

• Contextual document retrieval using RAG and ChromaDB

• Prompt & context management using MCP

Project Structure

.
├── agents/               # LLM agents (e.g. researcher, reviewer)
├── config/               # Configurations
├── db/                   # SQLite store
├── graphs/               # LangGraph workflow 
├── mcp/                  # Model Context Protocol (MCP) implementation
├── nodes/                # LangGraph nodes
│     └── conditions      # nodes conditions
├── rag/                  # RAG (retrieval-augmented generation) logic
├── state/                # Shared state classes for LangGraph workflows
├── tests/                # LangGraph test
├── .env.example          # Sample environment variables
├── .gitignore            
├── Makefile              # Task runner
├── requirements.txt      # Python dependencies
└── README.md             

Requirements

• Python=3.11.11

• Virtual environment (recommended)

• make (optional)

To run the project

Step 1:

Create and activate a virtual environment (recommended)

python -m venv .venv
source .venv/bin/activate     
# On Windows: .venv\Scripts\activate 

Step 2:

Option 1: Using Makefile

make setup

Option 2: Without Makefile

pip install -r requirements.txt

Step 3:

Copy the .env.example file and rename the file to .env

Step 4:

Add API keys to .env.

Usage

Step 1:

To run the MCP development server

Option 1: Using Makefile

make run-mcp

Option 2: Without Makefile

mcp dev mcp/server.py

Step 2:

• Visit http://localhost:5173 to the browser.

• Change the Command to python

• Change Arguments to mcp/server.py

• Click to Connect and wait for connection

• After establishing the connection, click Tools -> List Tools -> research

• Then write the research topic and Run Tool

To Test Graph Workflow

make test-graph # with make
python tests/test_graph.py # without make