MCP -App

This MCP Server App provides a complete MCP workflow with LangGraph agent integration. It wires together a FastMCP server implemented in Python with a LangGraph-enabled agent that can orchestrate multiple MCP tools. The core capabilities include dice rolling using natural language with the dice_roller tool, real-time web information retrieval via Tavily, and content creation for social posts using the social_content_creator tool. Additional integrations let you search GitHub repositories and browse repository contents through the github_tool, as well as generate presentation-ready images for slides. The design demonstrates how to compose multi-tool workflows (e.g., asking the agent to search the web, then create a social post and fetch a slide image) using a ReAct-style or StateGraph workflow pattern. To interact, run the server in STDIO transport mode and connect via the LangGraph client, or use the provided interactive demo scripts. The architecture enables agents to reason about tasks and execute them across tools, producing end-to-end results such as curated posts, image selections, or code/file fetches from GitHub.

Typical usage flows include:

• Asking the agent to roll dice with D&D notation and then generate a summary of the results.
• Requesting a web search for current information and immediately creating a social post that references the findings.
• Generating a quote card or image assets and then retrieving related GitHub repository content for further exploration.

Prerequisites

• Python 3.13+ installed on your system
• Access to an OpenAI API key for the LangGraph agent
• Tavily API key for web search (required)
• Optional: Unsplash API key for images (fallbacks to Lorem Picsum if not provided)
• Optional: GitHub Token to access private repositories

Installation steps

1. Clone the repository and set up the project

git clone <repository-url>
cd AIE7-MCP-Session

2. Create a local environment file (optional but recommended) and populate keys

cp .env.sample .env

Edit .env and set keys such as OPENAI_API_KEY, TAVILY_API_KEY, UNSPLASH_ACCESS_KEY, GITHUB_TOKEN as needed

3) Install dependencies
```bash
uv sync        # Recommended (via uvx)
# OR if you prefer pip:
pip install -r requirements.txt

4. Run the server and client for verification

uv run run_server.py        # Start the FastMCP server (STDIO transport used with LangGraph)
uv run run_client.py        # Start an interactive LangGraph client demo

5. Validate tests (optional)

uv run tests/test_mcp_integration.py
uv run tests/test_server.py

Notes

• Ensure your API keys are securely managed and not committed to source control.
• The STDIO transport demonstrates process-to-process communication between the LangGraph agent and the MCP server. Adjust the transport setting if you deploy in a different environment.

Environment and configuration tips:

• OPENAI_API_KEY: required for LangGraph agent reasoning with external tools.
• TAVILY_API_KEY: required for web search tool access.
• UNSPLASH_ACCESS_KEY: optional; used for image generation of social posts and slide images. If absent, the system may fallback to placeholder images.
• GITHUB_TOKEN: optional; enables private repository access and higher rate limits when using the GitHub tooling.

Common issues:

• If the MCP server fails to start, verify that Python 3.13+ is used and that uvx is installed and available in PATH.
• When running in STDIO mode, ensure the client and server processes are started in a coordinated manner to allow proper inter-process communication.
• If API keys become invalid or reach rate limits, regenerate keys and update the environment file.

Configuration options:

• You can extend mcp_config with additional servers or tool integrations by adding entries under mcpServers with the same command/args pattern.
• To customize behavior, modify the individual tool modules (dice_roller.py, social_content_creator.py, github_tool.py) to suit your workflow needs.