Code Execution & Sandboxes MCP servers

Runtime security proxy for MCP that blocks prompt injection, SSRF, shell/SQL injection and credential exfiltration, plus trust scores for npm MCP packages.

Interact with TUI/CLI applications through a structured terminal state tree with element detection, like Playwright for terminals.

Developer tools for file editing, shell command execution and screen capture.

The shell around your coding agent: a neuro-symbolic code graph with calibrated trust, via MCP.

Ephemeral data sandbox for AI workflows with guardrails and security

MCP server that gives AI agents (Claude Code, Cursor, Windsurf) real interactive terminal sessions — run REPLs, SSH, databases, Docker, and any interactive CLI with clean output and smart completion detection

Interactive PTY sessions for AI agents: local shells, persistent SSH sessions with attach/detach, and serial ports.

Sandboxed shell exec for MCP clients: run untrusted agent commands in a gVisor container.

Run code in multiple programming languages locally via Docker.

Approval-first local terminal control: policy-based command planning and approval, DeepSeek as an advisor, and signed execution receipts.

Mac window manager and menu bar utilities: arrange windows across displays, snap to custom zones, save workspaces, window screenshots and on-device OCR.

Inspect, debug and drive running WPF apps without source changes: visual tree, dependency properties, bindings, DataContext, screenshots, clicks and text input.

Drive native GUI apps under development: launch, screenshot, read the accessibility tree, inject input, tail logs and diff frames on desktop, Android and iOS.

Linux system administration via typed actions instead of shell strings, with a signed audit log, one-time approval receipts and automatic rollback.

Standalone MCP filesystem tools server — view, edit, shell, workdir.

YADE discrete element method (DEM) simulation: search API docs, run code via a REPL or background tasks, monitor progress and review task history.

GNOME desktop automation via D-Bus: screenshots, window management, mouse and keyboard input, clipboard, workspaces and notifications on GNOME 45 to 49.

Read-only desktop context through OpenWand: current selection, clipboard, active window, browser page and screen.

The terminal AI agents can drive: spawn panes, run commands, read screens, scrolling screenshots.

Accessibility-first computer use on macOS, Windows and Linux: element IDs instead of pixels, background input, a pointer overlay and its own Chrome tab group.

Agent-oriented programming language: agents emit a JSON AST that the compiler validates, type- and effect-checks, verifies with Z3/SMT and compiles to WASM.

Self-hosted SSH and Kubernetes access broker: ephemeral per-operation credentials the model never sees, per-command firewall, human approvals and audit log.

Arbitrary code execution and tool-use platform for LLMs by Riza

Launch, screenshot and interact with any TUI app.

Deterministic pre-execution filter for agent tool calls that blocks scope creep (credential reads, SSRF, destructive shell/SQL) and prompt injection.

Physical USB HID keyboard and mouse control via a Raspberry Pi Pico 2: move, click, drag, type, key combos and scroll, with a mock mode for testing.

Stateful, auditable shell sessions over local, SSH and Docker, with secret redaction, output budgeting and a read-only live transcript viewer.

SWI-Prolog execution for LLMs with CLP(FD), negation as failure and recursion.

A powerful CLI command execution MCP server that enables running shell commands with structured output

A powerful CLI command execution MCP server that enables running shell commands with structured output

MCP PTY server providing AI agents with real interactive terminal access

MCP server for interactive terminal sessions — SSH, REPLs, database CLIs, TUI apps

Local command wrapper for AI coding agents: runs shell commands, stores history locally and returns compressed terminal output to reduce context noise.

dwic — a product designer inside your terminal. MCP server + CLI that audits your design tokens + markup for WCAG contrast failures, mandated-accent drift, and structural gaps, then remembers findings across sessions.

SSH access with persistent sessions, SFTP and port forwarding, with secrets injected via stdin and redacted from output before reaching the LLM.

Execute code through the Piston remote code execution engine.

Execute bash commands with background job support.

An MBTI-powered terminal pet companion for Claude Code and OpenClaw.

Persistent Python sandbox for token-efficient codebase exploration in MCP clients

SSH control of remote Linux, macOS and Windows servers: line-level file editing, background tasks, sudo, host aliases, and directory and archive operations.

Search, retrieve and diff the output of past terminal commands, recorded locally with exit codes, cwd and duration, with secret redaction by default.

Uncensored AI coding agent for your terminal — Claude Code alternative with MCP support. Reads code, writes files, runs commands. Drives IDA Pro, Roblox Studio, Wireshark, Blender, and any MCP server. No refusal layer.

Run YAML-defined shell command stacks as MCP tools

Drive terminal coding agents by voice, with receipts that never claim more than they proved.

Remote server work via the local OpenSSH client: run commands, transfer files with checksums, search logs and audit hosts, with destructive commands blocked.

Persistent terminal MCP with one harness-based launcher for Claude Code, Codex CLI, Grok CLI, and Cursor Agent CLI, plus durable PTYs for SSH, containers, and REPLs.

Code-mode MCP server: collapses user-defined tools into code execution plus docs search, sandboxed in node:vm, WASM or remote microVMs.

WasmAgent MCP server — WASM-sandboxed code execution with capability manifests and Tasks API.