Agent VM

A container-based runtime for AI agents with an embedded MCP (Model Context Protocol) resolver. Agent VM gives your agent a sandboxed Linux environment with filesystem and shell access exposed through a standard MCP interface.

Quick Start

Build the base image

docker build -f images/base/Dockerfile -t agent-vm-base .

Run a container

# Interactive shell with workspace mount
docker run -it --rm -p 7777:7777 -v $(pwd)/my-project:/workspace agent-vm-base

# Detached — MCP resolver only
docker run -d --rm -p 7777:7777 -v $(pwd)/my-project:/workspace agent-vm-base sleep infinity

Verify the resolver is running

curl http://localhost:7777/health
# {"status":"ok"}

Connect an MCP client

The resolver speaks MCP Streamable HTTP. Point any compatible client at:

http://localhost:7777/mcp

See the MCP Resolver documentation for protocol details and tool reference.

Using Pre-built Images

docker pull ghcr.io/sean-lai-sh/agent-vm-base:latest

Documentation

• MCP Resolver Documentation
• Model Context Protocol

Python SDK

For provisioning and managing Agent VMs programmatically, see agent-vm-sdk.

License

MIT