# oraclemcp

> An MCP (Model Context Protocol) server that gives AI agents access to Oracle Database without the risk: read-only by default, a fail-closed SQL guard, OAuth 2.1, and a tamper-evident audit trail. Pure Rust, engine-free, stdio and streamable HTTP.

Source: https://durakovic.ai/projects/oraclemcp/ · Author: Dipl.-Ing. Muhammed Durakovic · Updated: 2026-10-14

- **Author:** [Dipl.-Ing. Muhammed Durakovic](https://durakovic.ai/about/)
- **Registry:** Listed in the [official MCP registry](https://registry.modelcontextprotocol.io/v0/servers?search=io.github.MuhDur/oraclemcp) as `io.github.MuhDur/oraclemcp`
- **Crate:** [`oraclemcp`](https://crates.io/crates/oraclemcp) on crates.io
- **Source:** [github.com/MuhDur/oraclemcp](https://github.com/MuhDur/oraclemcp)
- **Driver:** Built on the [original pure-Rust Oracle driver](https://durakovic.ai/projects/rust-oracledb/) — no Instant Client

## The problem

Connecting an LLM agent to a production database is easy. Making sure it can never drop a table, exfiltrate a schema or run an unbounded query is not. Most database MCP servers trust the model; oraclemcp does not.

## How it stays safe

- **Read-only by default** — write capabilities require explicit, confirmation-gated operating levels.
- **Fail-closed SQL guard** — statements that cannot be proven safe are rejected, not attempted.
- **Least privilege** — governed access per tool and per operating level.
- **OAuth 2.1** for remote use over streamable HTTP.
- **Tamper-evident audit trail** — every statement an agent runs can be reviewed afterwards.

[View the source on GitHub](https://github.com/MuhDur/oraclemcp)