langchain_mcp

This repository demonstrates a minimal working MCP (Multi-Server Control Plane) setup using LangChain, with:

• A dummy jobs and employee API (FastAPI)

• Two MCP servers (jobs and employee feedback, each on its own port)

• A Python client that can query both servers simultaneously (multi-server, multi-tool)

Requirements

• Python 3.9+

• pip

• Node.js (optional, only if you want to build a frontend)

• An OpenAI API key (for GPT-4o)

Related MCP server: Linear

Setup

1. Clone the repository

git clone https://github.com/nishant-Tiwari24/mcp.git
cd mcp

2. Create and activate a virtual environment

python3 -m venv .venv
source .venv/bin/activate

Note: .venv is gitignored. You must create it yourself.

3. Install Python dependencies

pip install --upgrade pip
pip install -r requirements.txt

4. Set your OpenAI API key

Create a .env file in the project root (not tracked by git):

OPENAI_API_KEY=sk-...your-key-here...

Or export it in your shell before running the client:

export OPENAI_API_KEY=sk-...your-key-here...

Running the Demo

1. Start the dummy jobs/employee API

uvicorn mcp_server.jobs_api:app --port 8001 --host 127.0.0.1

2. Start both MCP servers (in a new terminal)

python run_servers.py

This will start:

• Jobs server on port 8000

• Employee server on port 8002

Or, to run them manually in separate terminals:

python mcp_server/server.py jobs
python mcp_server/server.py employee

3. Run the multi-server client (in a new terminal)

python langchain_mcp_client.py

• The client will connect to both servers and can use tools from both in a single conversation.

• Make sure your OpenAI API key is exported or in a .env file.

File Structure

• langchain_mcp_client.py — Python client for querying both MCP servers (multi-server, multi-tool)

• mcp_server/server.py — MCP servers (jobs and employee feedback, each on its own port)

• mcp_server/jobs_api.py — Dummy FastAPI backend for jobs and employee data

• run_servers.py — Script to start both MCP servers at once

• requirements.txt — Python dependencies

• .gitignore — Excludes .venv, .env, and other environment files

Notes

• The .venv directory and .env file are not included in the repo. You must create them locally.

• Only the minimal, required files are tracked in git.

• If you want to add a frontend, you can do so separately (not included in this repo).

Example Usage

• Multi-server client:

  • Query: "I need to find similar jobs for an AI engineer position in San Jose, CA with 4-5 years of experience, and also get a feedback summary for Kalyan P. Can you help me with both?"

  • The client will use both tools: find_similar_jobs and summarize_employee_feedback.

License

MIT