Flitzpiepe93 Mcp Server

MCP Database Server for AI Agents

> Proof of concept. A time-boxed exploration of secure agent–data > boundaries — designed to run locally, not a production service. Cloud > deployment and agent lifecycle are designed in the > docs but intentionally left > unimplemented; the PoC's job is to prove the boundaries work end to end.

An MCP server that gives AI agents controlled, audited access to a database through business-level tools instead of raw SQL. Every call is authenticated, authorized against tool-level scopes, and logged — so you always know which agent read what, and when.

The public Titanic dataset (SQLite) stands in for sensitive data (think insurance records): that sensitivity is what motivates the authentication, access control, and auditability. The one implemented tool, get_survival_rate, returns survival figures grouped by passenger class or sex.

Run it locally

Requirements: Docker (incl. Compose v2) and make.

cp .env.example .env   # local config (dev defaults, no real secrets)
make up                # start Keycloak + MCP server
make run-client        # run the example agent once against the server

The client fetches a token from Keycloak, calls the tool, and prints the result — while the server's audit middleware logs every call with agent, tool, and parameters. Run make with no argument to list every command.

Repository layout

server/     MCP server: the tool, auth, audit middleware, and the
            repository layer that isolates it from the database
client/     Example agent that fetches a token and calls the tool once
keycloak/   Realm export (client, scopes, audience mapper) imported on startup
data/       Checked-in SQLite Titanic dataset, so the demo needs zero setup
docs/       Full design write-up, published to GitHub Pages via mkdocs

Inside server/src/server/, app.py wires everything together, auth.py and audit.py cover authentication and the audit trail, and repository/ holds the swappable data-access layer (sql.py for real use, memory.py for tests). Root-level files (docker-compose.yml, Makefile, .env.example) orchestrate the local stack.

Dataset

The Titanic dataset (891 passengers) comes from seaborn's built-in titanic dataset (BSD-licensed). It is checked in as a small SQLite file (data/titanic.db) so the demo runs with zero setup.

---

Originally built as a time-boxed coding challenge, then reworked into this portfolio piece.

Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: flitzpiepe93
• Source: flitzpiepe93/mcp-server
• License: Apache-2.0
• Homepage: https://flitzpiepe93.github.io/mcp-server/

Install and usage instructions live in the source repository linked above.