MCP Hub
Back to servers

Expense Tracker MCP Server

Enables users to manage expenses with category support, including adding expenses with date, amount, category, and notes, using SQLite for persistence.

glama
Updated
May 7, 2026

Remote MCP Server and Client Showcase

Python MCP Streamlit Manim

A practical Model Context Protocol workspace with two connected parts:

  • A remote-ready FastMCP expense server for structured expense tracking.
  • A Streamlit MCP client app with local math tools, web search, Groq/OpenAI LLM support, and Manim animation rendering.

This repo is designed as a compact reference for building MCP tools, exposing them through a server, and consuming them from a chat UI.

What This Shows

AreaWhat is included
MCP serverFastMCP tools for expense management with SQLite persistence
MCP clientStreamlit chat app using langchain-mcp-adapters
Tool callingMath tools, Tavily-backed search, and Manim video rendering
LLM providersGroq by default, OpenAI available through config
Local testingSmoke tests for MCP tools, search backend, and Manim rendering

Repository Structure

test-remote-mcp-server/
|-- main.py                  # FastMCP expense server
|-- proxy.py                 # Proxy entrypoint for remote access
|-- categories.json          # Expense category definitions
|-- pyproject.toml           # Server dependencies
|-- uv.lock                  # Server lockfile
|-- mcp-client-app/          # Streamlit MCP client showcase
|   |-- client2.py           # Main web app
|   |-- main.py              # Local math MCP server
|   |-- test_tools.py        # Tool smoke tests
|   |-- manim_test_scene.py  # Direct Manim render test
|   |-- .env.example         # Safe environment template
|   `-- README.md            # Client app guide
`-- README.md

FastMCP Expense Server

The root server exposes expense tracking tools backed by SQLite. It is useful for testing remote MCP tool workflows and structured tool arguments.

Server Setup

git clone https://github.com/deepakbishnoi717/test-remote-mcp-server.git
cd test-remote-mcp-server
uv sync

Run the local server:

uv run python main.py

Run the proxy entrypoint:

uv run python proxy.py

Expense Tool Example

Use natural language from an MCP-compatible client:

Add an expense for 450 INR in Food, subcategory Lunch, with note "team meal".

Streamlit MCP Client App

The client app lives in mcp-client-app/. It demonstrates a chat UI that can call tools from both an MCP server and direct LangChain tools.

Highlights:

  • Local MCP math server: add, subtract, multiply, divide
  • Tavily-backed web search tool exposed as brave_search
  • Manim renderer exposed as render_manim_code
  • Groq model support by default
  • OpenAI fallback through .env
  • Windows-friendly launcher: run_app.bat

Start the client:

cd mcp-client-app
copy .env.example .env
run_app.bat

Open:

http://localhost:8501

Demo Prompts

Try these in the Streamlit app:

Use the math tool to multiply 12 by 8, then subtract 10.
Search the web for the latest Model Context Protocol updates and summarize them.
Use render_manim_code to create a Manim animation of a blue circle transforming into a green square. Return the rendered video path.

Validation

From the client folder:

.\.venv\Scripts\python.exe -B test_tools.py

Direct Manim render test:

.\.venv\Scripts\python.exe -B -m manim -ql manim_test_scene.py GeneratedScene --media_dir manim_outputs\direct_media

Expected Manim output:

manim_outputs\direct_media\videos\manim_test_scene\480p15\GeneratedScene.mp4

Environment Variables

The client app uses .env.example as a safe template:

VariablePurpose
GROQ_API_KEYRequired for Groq chat models
TAVILY_API_KEYRequired for web search
OPENAI_API_KEYOptional OpenAI provider
LLM_PROVIDERgroq or openai
GROQ_MODELDefault Groq model
OPENAI_MODELDefault OpenAI model

Never commit a real .env file.

Tech Stack

  • FastMCP and MCP
  • LangChain tool binding
  • Streamlit
  • Groq and OpenAI chat providers
  • Tavily Search API
  • Manim Community
  • SQLite
  • uv

Notes

  • The root server and the client app are intentionally separated so each can be studied or deployed independently.
  • Generated videos and local virtual environments are ignored by Git.
  • The client defaults to Groq to avoid OpenAI quota errors during local testing.

Reviews

No reviews yet

Sign in to write a review