Skip to main content
Glama
Command Center
by 1xOps

1xOps Command Center

Your AI agent keeps a terminal open — tests, logs, REPLs and SSH — inside the folders you allow.

A local MCP server: shell sessions that stay open between turns, plus text-file tools, under one folder limit and one command blocklist that only a person can change. It runs on your machine with your permissions and sends nothing anywhere.

/plugin marketplace add 1xOps/command-center
/plugin install command-center@command-center

💼 What teams use it for

Scenario

How it works

Why it helps

🧪 Fix a failing test suite

Run the tests, read the failures, edit, run again — in the same shell

The environment stays set up; the engineer reviews one diff

🚨 Watch logs during an incident

Start a tail, read new output on demand, stop it with force_terminate

A second pair of eyes on live output; nothing left running afterwards

📊 Check data in a REPL

python3 -i or node -i keeps the data loaded across turns

Answers come from the data, not from a script re-run each time

🔐 Diagnose a server over SSH

One ssh session kept open across turns

Faster remote triage. The blocklist checks the remote command and each typed line; the remote account's permissions are still the real limit

🏢 One approved setup for the team

Commit a pinned .mcp.json

Every seat runs the same reviewed version

Related MCP server: TerminalMCP

🧩 Clients and platforms

Client

Install

macOS

Linux

Windows

Claude Code — plugin (server + 6 skills)

/plugin install above

✓

✓

✓ ¹

Claude Code — server only

claude mcp add

✓

✓

✓ ¹

Claude Desktop

.mcpb from Releases

✓

✓ ¹

OpenAI Codex CLI

codex mcp add

✓

✓

✓ ¹

VS Code (agent mode)

.vscode/mcp.json

✓

✓

✓ ¹

Node.js 24 or later (Claude Desktop installs it from the .mcpb). 20 tools — see mcp-server/README.md. ¹ Windows code paths and tests exist; a Windows CI run has not happened yet.

Replace /path/to/project with the folders the agent may use; they become allowedDirectories for that session.

The plugin passes no folder. With no folder and an empty allowedDirectories there is no folder limit. Set allowedDirectories in ~/.command-center/config.json, or use a server-only install with a folder.

# Claude Code, server only
claude mcp add command-center -- npx -y @1xops/command-center@0.4.2 /path/to/project

# OpenAI Codex CLI
codex mcp add command-center -- npx -y @1xops/command-center@0.4.2 /path/to/project

VS Code — .vscode/mcp.json:

{ "servers": { "command-center": { "type": "stdio", "command": "npx", "args": ["-y", "@1xops/command-center@0.4.2", "${workspaceFolder}"] } } }

Claude Desktop — open command-center-0.4.2.mcpb from Releases; it asks which folders Command Center may use.

Claude Code — commit .mcp.json at the project root:

{ "mcpServers": { "command-center": { "command": "npx", "args": ["-y", "@1xops/command-center@0.4.2", "."] } } }

Codex — ~/.codex/config.toml:

[mcp_servers.command-center]
command = "npx"
args = ["-y", "@1xops/command-center@0.4.2", "/path/to/project"]

Check where a build came from: npm view @1xops/command-center@0.4.2 dist.attestations (npm provenance from the CI release).

Non-root user; sees only the mounted folder. Build from a checkout of this repo:

docker build -t command-center:0.4.2 mcp-server
claude mcp add command-center -- docker run -i --rm --network none -v "/path/to/project:/workspace" command-center:0.4.2

--network none cuts network access for untrusted work; drop it if sessions need the network.

Command Center runs commands with your permissions and trusts the AI client driving it. It does not detect prompt injection.

Control

What it does

Security boundary?

allowedDirectories

Every file tool and session working folder must be inside these folders (symlinks resolved first). Empty = no limit

No

blockedCommands

Refuses listed programs wherever they start a command, incl. after ; && | $(…), sh -c, env/timeout wrappers; always refuses eval / Invoke-Expression. Defaults include sudo, dd, mount, diskpart, reg

No

Person-only settings

set_config_value refuses allowedDirectories, blockedCommands, defaultShell; file tools refuse the config folder

No

Container

Non-root, only the mounted folder visible

Yes, for the host filesystem

Limits: a program started in a session can reach any path and network you can, and could edit the config file (it applies at the next start). The command check reads command lines (including lines typed into a shell or ssh session), not scripts or REPL input. Each session keeps its last 1 MiB of output in memory.

Privacy: no telemetry, analytics, update checks or remote calls. See SECURITY.md · PRIVACY.md.

📁 In this repository

Folder

What it is

mcp-server/

npm package @1xops/command-center: server, tests, Claude Desktop bundle, container

plugins/command-center/

Claude Code plugin: the pinned server plus six skills

docs/supply-chain/register.md

Every shipped package and its licence, signed by the owner

MCP Registry io.github.1xOps/command-center · Apache-2.0 © 2026 1xOps · Report vulnerabilities privately: SECURITY.md

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      A
      quality
      B
      maintenance
      Enables an AI agent to take full control of a machine — running shell commands, editing files, managing git, packages and processes, and reaching the network — with batched execution, server-side variables and selectable tool profiles that keep token usage low. It also speaks HTTP so the same agent can operate a remote build box, NAS or VPS.
      26
      1
      MIT
    • A
      license
      A
      quality
      B
      maintenance
      Enables AI assistants to spawn and control persistent local shells, SSH sessions, REPLs, debuggers, and serial/UART devices with zero-loss output capture, timestamped history, and atomic expect matching for command execution and prompt synchronization.
      12
      MIT
    • A
      license
      A
      quality
      B
      maintenance
      For people who want their AI to run the build, read the error and check the dev server itself, instead of pasting output into chat. Works with Claude Desktop (one-click installer), Cursor, Claude Code and Cline. Runs locally over stdio or HTTP, no telemetry. A seatbelt for everyday mistakes, not a sandbox.
      20
      91 npm
      MIT