Mcpod

mcpod (MCP + Pod) is a command-line tool for installing, managing, and running Model Context Protocol (MCP) servers inside isolated Docker containers.

This project was built for the 2026 UCF Bloomknights 12-hour hackathon. The goal is to make MCP servers easier to try, safer to run, and simpler to connect to popular AI development clients without giving every server unrestricted access to your machine.

Why mcpod?

MCP servers can be powerful: they expose tools, data sources, and local capabilities to AI assistants. That also means they deserve a careful runtime boundary. mcpod treats each server like an app with an explicit permission request:

• no network access unless the server's config.mcpod asks for it,
• no filesystem access unless the server's config.mcpod asks for it,
• bind mounts are scoped and read-only by default;

Features

• 📦 Containerized MCP servers — run installed MCP servers as Docker containers instead of directly on your host.
• 🔎 Marketplace browsing — search, list, inspect, and fetch server configs from an MCP server registry.
• 🛠️ One-command installs — install marketplace entries, local paths, or Git repositories that provide a config.mcpod file.
• 🛡️ Permission-first runtime — enforce declared network, filesystem, and compute permissions when containers are created.
• 🔄 Client registration — register installed servers with Claude Code, Claude Desktop, VS Code, Cursor, or LM Studio.
• 🕒 Lifecycle management — start, stop, remove, and update installed servers from the CLI.
• 🤖 Automatic Config Generation — --interpret-unsafe can use Gemini to generate a config.mcpod from docs when a server does not provide one, but only after the user explicitly opts in.

Web marketplace

Alongside the CLI, mcpod ships a web frontend for discovering servers before you install them. The landing page pitches the sandbox-by-default model and shows the install flow in action:

Browse the registry, filter by capability, name, or publisher, and see every available pod at a glance:

Open any server to inspect its overview, requested permissions, and the exact config.mcpod manifest the registry serves — plus a one-line install command and ready-to-copy client config:

Requirements

• Node.js 26
• npm
• Docker running locally for container operations
• Optional: Gemini API key in GEMINI_API_KEY for mcpod install --interpret-unsafe

Getting started

Clone the repository and install workspace dependencies:

git clone 
cd mcpod
npm install

Run the CLI locally without linking:

node cli/bin/mcpod.js --help

Or link the package during development so mcpod is available on your PATH:

cd cli
npm link
mcpod --help

CLI usage

mcpod [command]

Available top-level commands:

mcpod marketplace search 
mcpod marketplace list
mcpod marketplace info 
mcpod marketplace fetch 

mcpod install [options] 
mcpod run [options] 
mcpod stop [options] 
mcpod rm [options] 
mcpod update [options] [name]

Browse the marketplace

mcpod marketplace search github
mcpod marketplace list
mcpod marketplace info my-server
mcpod marketplace fetch my-server

Install a server

Install a server and review its requested permissions interactively:

mcpod install my-server

Every install surfaces the server's requested network, filesystem, and compute permissions before anything is pulled:

After you grant permissions, mcpod detects which MCP clients are present and lets you pick where to register the server:

Once registration finishes, the server is installed and ready to run:

Accept requested permissions non-interactively:

mcpod install --yes my-server

Register the installed server with an MCP client:

mcpod install my-server --client claude-code:project
mcpod install my-server --client vscode:global --client cursor:project

Supported client specs include:

• claude-code:project
• claude-code:global
• claude-desktop:global
• vscode:project
• vscode:global
• cursor:project
• cursor:global
• lm-studio:global

Run and stop servers

mcpod run my-server
mcpod run my-server --cwd /path/to/project
mcpod stop my-server
mcpod stop --all
mcpod stop --all --force

Running a server creates its container with only the declared permissions and starts it over the configured transport:

Remove or update servers

mcpod rm my-server
mcpod rm my-server --force
mcpod update my-server
mcpod update --all

config.mcpod

MCPod servers are described by a YAML file named config.mcpod. It declares where the server source comes from, how to run it, what environment values it needs, and which permissions the container may receive.

A minimal config looks like this:

metadata:
    name: my-server
    description: "Example MCP server"
    version: "1.0.0"

source:
    url: "https://github.com/example/my-server.git"

image: node:26-alpine
command: ["node", "server.js"]
transport: stdio

permissions:
    network:
        outbound: false
    filesystem: []

See the repository's [config.mcpod](./config.mcpod) for a more detailed example covering environment prompts, secrets, ports, compute limits, and lifecycle scripts.

Architecture

MCPod has two main parts:

1. MCPod CLI (cli/) — the user-facing command-line app. It installs servers, registers client config entries, creates Docker containers with Dockerode, starts and stops server containers, and stores install state under ~/.mcpod/.
2. Marketplace Server (marketplace/) — an Express-based registry service that serves a paged JSON index of available MCP servers and their config.mcpod files.

The CLI is intentionally a thin orchestrator. Marketplace data and server configs are treated as untrusted input and must be validated before they influence container creation.

License

This project is licensed under the terms in [LICENSE](./LICENSE).

Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: skyclo
• Source: skyclo/mcpod
• License: MIT
• Homepage: https://devpost.com/software/mcpod

Install and usage instructions live in the source repository linked above.