FastAPI + MCP + LangGraph Demo

This is a very small learning project.

One app exposes the same note operations in two ways:

• REST API with FastAPI

• MCP tools for AI clients and agents

What is included

• POST /notes to create a note

• GET /notes/{note_id} to read a note

• GET /graph-rag?question=... for a tiny Graph RAG example

• MCP tools:

  • create_note

  • get_note

  • graph_rag_search

• mcp_client_demo.py to call the MCP server directly

• langgraph_agent_demo.py to give a LangGraph agent access to the MCP tools

• graph_rag_demo.py to show a tiny knowledge-graph retrieval flow

Related MCP server: flomo-mcp

Install

.venv/bin/pip install -e .

Run the server

.venv/bin/uvicorn main:app --reload

Open:

• FastAPI docs: http://127.0.0.1:8000/docs

• MCP endpoint: http://127.0.0.1:8000/mcp/

Try the REST API

Create:

curl -X POST http://127.0.0.1:8000/notes \
  -H "content-type: application/json" \
  -d '{"title":"hello","content":"from fastapi"}'

Read:

curl http://127.0.0.1:8000/notes/1

Try the MCP client

Start the server first, then run:

.venv/bin/python mcp_client_demo.py

Try the Graph RAG demo

Run the standalone script:

.venv/bin/python graph_rag_demo.py

Or hit the REST endpoint:

curl "http://127.0.0.1:8000/graph-rag?question=How%20is%20Bob%20connected%20to%20Neo4j%3F"

What the Graph RAG demo is showing

This part is intentionally tiny and concept-focused:

• a few short source documents

• a few graph triples like Bob -> owns -> Search API

• retrieval over graph structure instead of only plain text matching

The most important concept is multi-hop retrieval. For example:

• Bob -> owns -> Search API

• Search API -> depends_on -> Graph RAG Service

• Graph RAG Service -> uses -> Neo4j

That lets the demo answer a question like How is Bob connected to Neo4j? by walking the graph path.

Try the LangGraph agent

Start the server first, then set your OpenAI key:

export OPENAI_API_KEY=your_key_here

Optional model override:

export OPENAI_MODEL=gpt-4.1-mini

Run the demo:

.venv/bin/python langgraph_agent_demo.py

Without OPENAI_API_KEY, the script will still show that LangGraph can load the MCP tools, but it will skip the actual agent run.

Notes

• Data is stored only in memory.

• Restarting the server clears all notes.

• This is intentionally minimal for learning.