Skip to content
Osaurus logo

Osaurus

Verified

Native macOS harness for AI agents with memory and offline execution.

MCP ServerFinance & CryptoLocal (stdio) 5.8k
View on GitHub
Updated 2026-06-15

What is the Osaurus MCP server?

Osaurus runs locally on Apple Silicon and keeps all context, memory, and tools on the user's Mac. It enables agents that remember prior tasks, execute autonomously, and connect to cloud providers only when chosen.

The system includes a sandbox powered by Apple's Containerization framework for safe shell and code execution without risking the host machine.

Install & connect

Set up this server, then add it to your MCP client.

Full setup instructions are in the GitHub repository.

Example prompts

Once connected, try asking your AI client:

Start a new coding agent with sandbox enabled and a working folder at ~/projects
List all my current agents and their memory summaries
Run the agent loop on my latest research notes and generate a todo list
Check status and confirm the server is running offline

Security & permissions

Runs locally via stdio on macOS with no remote access unless user explicitly connects cloud models; all memory and secrets stay on the device.

What you can do with Osaurus

Autonomous Research Agent

Create a dedicated research agent that maintains memory across sessions and uses file, search, and git tools within a chosen working folder.

Secure Code Execution

Run code in an isolated Linux VM sandbox with Python, Node.js, and compilers while keeping all data on-device.

Personal AI Identity

Maintain a cryptographic identity and encrypted agent database that persists across restarts without sending data to external servers.

How to use Osaurus

  1. 1Install via brew install --cask osaurus or download the DMG
  2. 2Launch Osaurus from Spotlight or run osaurus ui in terminal
  3. 3Start the server with osaurus serve
  4. 4Create agents and assign working folders or enable sandbox
  5. 5Connect compatible models or use local Apple Foundation Models

Osaurus: pros & cons

Pros

  • +Fully offline operation with local models on Apple Silicon
  • +Native Swift implementation without Electron overhead
  • +Isolated Linux VM sandbox for safe autonomous execution
  • +Automatic RAG-based tool selection for agents

Cons

  • Requires recent macOS and Apple Silicon hardware
  • Sandbox features need macOS 26 or later
  • No pre-defined tool list exposed for MCP clients
Did you find this helpful?

Frequently asked questions

No, it works fully offline with local models while allowing optional cloud connections.

User reviews

Verified reviews from the community shape this listing's rating.

Loading reviews…

Sign in to review

Promote Osaurus

Add this badge to your website, or share the tool.

DFeatured on DhanasviOsaurus 1