GetMe MCP Server

A High-Performance Key-Value Store

📑 Index

• Overview
• Project Structure
• Core Architecture
• Getting Started
  • Running the Server
  • Using the CLI
  • HTTP Proxy
  • MCP Server
• Running Benchmarks & Tests
• SDKs
• License

📖 Overview

getMe is a persistent, embeddable key-value store written in Go. It is inspired by the design of Bitcask and is optimized for high write throughput and low-latency reads.

It uses a log-structured storage approach, ensuring that all data is appended sequentially. It uses Unix Domain Sockets (UDS) for incredibly fast local inter-process communication, alongside several interfaces like an HTTP proxy, a CLI, and a Model Context Protocol (MCP) server for LLMs.

🏗 Project Structure

This project is a monorepo containing the core storage server, multiple client interfaces, and tools.

• server/: The core storage daemon and engine. Implements the log-structured hash table for persistent storage. See server/README.md for architectural deep-dives.
• cli/: A command-line interface for interacting with the getMe server for testing and debugging.
• sdks/: Client libraries (goSdk, javaSdk, jsSdk, pythonSdk) to integrate getMe into your applications.
• http-proxy-go/: An HTTP server built using the goSdk that exposes the core engine's Unix Domain Socket connection over standard HTTP routes.
• mcp-server/: A Model Context Protocol (MCP) server that exposes the getMe database as tools to Large Language Models (like Claude or Cursor).
• commons/: Shared code, socket paths, types, and constants used across the monorepo to ensure consistency.
• utils/: Shared utility packages, including logging stack configurations (Loki + Alloy + Grafana).

Spotlight: The curated inner docs are the quickest way to understand the system end-to-end. Start with server/README.md for architecture fundamentals, then explore the cli and mcp-server modules for integrations.

🧠 Core Architecture

The storage engine relies on a few core principles:

• Log-Structured Storage: All data is written to an append-only log file. This makes writes extremely fast as it avoids slow, random disk I/O.
• In-Memory Hash Index: A hash table is kept in memory, mapping each key to the exact location of its value on disk. This allows for very fast read operations (typically one disk seek).
• Compaction: A background process that periodically cleans up old, stale data from the log files to reclaim disk space.
• Fast Local Transport: Communication is done predominantly via Unix Domain Sockets, avoiding standard TCP overhead locally.

🚀 Getting Started

Running the Server

The repository ships with helper scripts to bootstrap the environment.

Option A: Local binaries + logging stack

Switch to the server module and run the local init script:

cd server
./init-server-local.sh

This script builds the Go binary into server/dist/, prepares data/log/socket directories, and starts the Loki + Alloy + Grafana logging stack via Docker Compose before launching the server in the foreground.

Warning: Do not prefix this script with sudo. It will invoke elevated privileges internally where needed. Using sudo at the top level causes permission errors for local development.

Option B: Full Docker Compose stack

From the same server directory run:

cd server
./init-server-docker.sh

This ensures host directories exist, exports your UID/GID, and invokes docker compose up --build to run everything in containers.

Using the CLI

Interact directly with the local server:

cd cli
go run . put mykey "hello world"
go run . get mykey
go run . delete mykey

HTTP Proxy

If you want standard HTTP REST endpoints instead of Unix Sockets, run the Go HTTP proxy:

cd http-proxy-go
go run main.go -port 8080

This will allow you to run curl http://localhost:8080/get?key=mykey.

MCP Server

getMe can be used by LLM clients (like Claude Desktop) through the Model Context Protocol.

cd mcp-server
uv run getme-mcp-server

(See mcp-server/README.md for configuration and integration instructions).

📊 Running Benchmarks & Tests

To ensure no performance regressions or to stress test the database:

1. Navigate to the specific module (e.g., server).
2. Run standard tests:
   bash

   Copy

   go test ./...
   

3. Run benchmarks:
   bash

   Copy

   go test -bench . ./...
   

   (Note: For heavier stress/correctness testing, look into server/tests/).

📦 SDKs

SDKs are available across different languages. Find them in the sdks/ directory:

• Go SDK
• JavaScript / TypeScript SDK
• Python SDK
• Java SDK

All SDKs interface directly with the Unix Domain Socket to provide optimal latency.

⚠️ Note on SDK Releases: SDK versioning and publishing is managed automatically via an Ephemeral Release Structure. The CI/CD pipelines autonomously orchestrate the entire release lifecycle—from creating detached commits and tagging them, to generating changelogs and pushing builds to public registries—all from a single bump-type trigger. This keeps the main branch entirely clean of meaningless version-bump commits. If you are exploring the code or contributing, do not manually bump versions in PRs. You can read more about this advanced architecture in the SDKs README.

📄 License

This project is licensed under the GNU Affero General Public License v3.0 (AGPLv3) - see the LICENSE file for details.