local-shell-mcp

local-shell-mcp addresses a common challenge: MCP clients like ChatGPT need to actually run commands, read and write files, drive a browser, or reach remote machines, but exposing the host directly is risky. It moves the execution boundary into a container or VM and gives the model a dedicated workspace, so real operations happen in a controlled environment with full auditability.

Core capabilities span several layers: persistent shell sessions can run tests, build projects, inspect logs, and debug; file tools support reading, writing, patching, searching, and code review; Git works through the standard CLI without a second incomplete abstraction; browser automation can extract page text, capture PNG or PDF evidence, or run full Playwright scripts; remote worker mode can control NAT, firewall, HPC, NPU, or lab machines that can only connect outward. It also provides Agent Skills discovery and loading, OAuth 2.1 and MCP App support for the ChatGPT connector, and an optional Goal plan mode so long tasks can continue across sessions.

It suits developers, operators, and AI automation enthusiasts who need to perform real development, debugging, testing, and operations inside an AI client, as well as teams that need to manage remote execution environments safely. Workspace isolation, command timeouts, output limits, environment filtering, secret scanning, and audit logs make automated operations more controllable.