ServerFS

•
6
•

Exposes explicitly configured Linux directories to AI agents with granular read/write control via a secure Model Context Protocol tunnel.

ServerFS is a secure Model Context Protocol (MCP) server designed to provide AI agents with controlled access to your Linux filesystem. By default, all exposed directories are read-only, with an opt-in mechanism for per-workdir file mutations. Agents interact with your server through the OpenAI Secure MCP Tunnel, enabling them to list, find, search, read, and stat files. For designated read-write workdirs, agents can also create, edit, and delete files using five narrowly defined, revision-guarded tools. The architecture ensures high security, with the MCP server container having no internet egress, no published ports, and running entirely isolated, making it a safe way to integrate AI agent capabilities with your server's filesystem.

主な機能

01Read-only file access by default with opt-in per-workdir write mutations.
02Secure exposure of Linux directories via OpenAI Secure MCP Tunnel.
03Granular file system tools: list, find, search, read, stat, create, edit, delete files and directories.
04Robust defense-in-depth security model including optimistic concurrency, atomic operations, and extensive deny rules.
05Containerized deployment using Docker and Docker Compose for easy setup and isolation.
060 GitHub stars

ユースケース

01Enabling AI agents to safely explore and analyze codebase structures on a server.
02Providing a controlled sandbox for AI agents to generate or modify project documentation.
03Automating specific file system operations with AI while maintaining strict security and auditability.