MCP Research Agent

A LangGraph-powered research assistant that uses three MCP servers to search the internet and fetch full webpage content, then formats a clean markdown report.

Built to demonstrate:

• Creating custom MCP servers with FastMCP
• Connecting multiple MCP servers via MultiServerMCPClient
• LangGraph agent graph with local memory (MemorySaver)
• LangGraph Studio / Docker graph visualization
• LangChain OpenAI model integration

Demo

https://github.com/user-attachments/assets/305ef919-6daa-4570-8b78-3a65fa78a186

MCP Servers

| Server | File | Transport | Tools | |---|---|---|---| | DuckDuckGo Search | servers/search_server.py | stdio | search_web | | Webpage Fetcher | servers/fetch_server.py | stdio | fetch_page | | Wikipedia search | servers/wikipedia_server.py | stdio | search_wikipedia |

Local Tools

| Tool | File | Purpose | |---|---|---| | format_report | app/tools.py | Formats findings into clean markdown |

Memory

Uses MemorySaver (in-process) keyed by thread_id. Each session maintains its full conversation history locally.

Project Structure

mcp-research-agent/
├── app/
│   ├── __init__.py
│   ├── graph.py          ← LangGraph agent graph + CLI entrypoint
│   ├── prompts.py        ← System prompt
│   └── tools.py          ← Local LangChain tools
├── servers/
│   ├── __init__.py
|   ├── wikipedia_server.py  ← MCP Server 1: Wiki search
│   ├── search_server.py  ← MCP Server 2: DuckDuckGo search
│   └── fetch_server.py   ← MCP Server 3: Webpage fetcher

├── .env                  ← your keys
├── .env.example
├── langgraph.json        ← LangGraph Studio config
├── requirements.txt
├── streamlit_app.py      ← Streamlit app entrypoint
└── README.md

Setup & Installation

1. Create a virtual environment

python -m venv .venv
source .venv/bin/activate      # macOS/Linux
# .venv\Scripts\activate       # Windows

2. Install dependencies

pip install -r requirements.txt

3. Add your OpenAI API key or local model

cp .env.example .env
# open .env and set OPENAI_API_KEY=sk-...

5. Run the agent in the terminal

python -m app.graph

or streamlit :

streamlit run streamlit_app.py

6. Visualize with LangGraph Studio (Docker)

Install the LangGraph CLI, then Start the local LangGraph development server:

pip install langgraph-cli

langgraph dev

Open the Studio URL shown in the terminal (usually http://localhost:8123). You will see the research_agent graph with all nodes, edges, thread history, and state at every step.

Example Questions

• What is MCP in LangChain and how does it work?
• Summarize the latest LangGraph features
• What are the latest advancements in AI agents in 2026?
• What are the differences between LangGraph and LangChain agents?

Key concepts this project demonstrates

| Concept | Where | |---|---| | MCP server creation with FastMCP | servers/ | | Multiple MCP servers via MultiServerMCPClient | app/graph.py | | LangGraph StateGraph with nodes + edges | app/graph.py | | Conditional routing with tools_condition | app/graph.py | | Local memory with MemorySaver | app/graph.py | | LangChain OpenAI model binding | app/graph.py | | Local LangChain tool alongside MCP tools | app/tools.py | | LangGraph Studio visualization config | langgraph.json |