openfactory（openfactory 测试）

OpenFactory Skill

Build custom Linux ISOs, deploy VMs, run compliance tests, and manage infrastructure — all through natural language.

Connection

OpenFactory exposes an MCP (Model Context Protocol) server over SSE.

Endpoint: https://build.openfactory.tech/api/mcp/sse

Authentication (pick one):

• API Key (recommended): Generate at the OpenFactory console under Settings > MCP Keys. Pass as api_key parameter on every tool call, or as Authorization: Bearer of_mcp_<key> HTTP header.
• Session Token: The first tool call returns a session_token in its response. Pass it to all subsequent calls for session continuity.

Session token is critical — cloud MCP clients rotate IPs between requests. Without passing session_token back, each call looks like a new anonymous user.

Tools

Builds

Recipes

Tests

VMs

Recipe Schema

{
  "name": "My Custom Server",
  "base_image": "ubuntu-24.04",
  "description": "What this variant is for",
  "features": ["ssh", "docker", "monitoring"],
  "packages": ["curl", "htop", "vim"],
  "users": [
    {"username": "admin", "password": "changeme", "groups": ["sudo"]}
  ],
  "services": [
    {"name": "ssh", "config": {"port": 22}}
  ],
  "security": {"hardening_level": "standard"},
  "hardware": {"min_memory_gb": 4, "min_storage_gb": 32},
  "networking": {
    "dns_servers": ["8.8.8.8", "1.1.1.1"]
  }
}

Supported base images: debian-bookworm, debian-trixie, ubuntu-24.04, fedora-40, fedora-43, opensuse-tumbleweed, elster-os

Common features: ssh, docker, desktop, development-tools, python, nodejs, monitoring, security-hardening, audit-logging, gxp, hipaa, ollama, claude-code, openclaw, nspawn-containers, agent-containers

Workflows

Build a custom Linux ISO

1. create_build(recipe)          → returns build_id
2. get_build_status(build_id)    → poll until status is "built"
3. run_tests(build_id)           → returns run_id
4. get_test_results(run_id)      → check passed/failed
5. get_iso_download_url(build_id) → download the ISO

Deploy a git repo as a service

1. deploy_from_git("https://github.com/user/app.git", base_image="ubuntu-24.04")
   → builds ISO with app auto-deployed at boot
2. get_build_status(build_id) → poll until built
3. create_vm(build_id)        → boot a VM running your app

Start from a template

1. list_recipe_categories()
2. list_recipes(category="healthcare")
3. create_recipe_from_template("gxp-medtech-station", "My MedTech", {
     "add_features": ["monitoring"],
     "add_packages": ["prometheus-node-exporter"]
   })
4. create_build(modified_recipe)

Test a build with compliance checks

1. run_tests(build_id, tests=["boot", "login", "network"], keep_environment=true)
2. get_test_results(run_id)
3. screenshot_vm(vm_name)  → visually verify the desktop

Error Handling

• Build failed: Use retry_build(build_id) or check the recipe for issues with validate_recipe(recipe).
• Test failed: Use get_test_results(run_id) to see which assertions failed and why.
• VM won't start: Check that the build completed successfully with get_build_status(build_id).

Rate Limits

• Builds: 1 concurrent per user (queued if busy)
• VMs: Up to 10 per user
• Tests: 1 concurrent per build
• API calls: No hard limit, be reasonable

Examples

"Build me an Ubuntu server with Docker and SSH"

{
  "name": "Docker Server",
  "base_image": "ubuntu-24.04",
  "features": ["ssh", "docker"],
  "users": [{"username": "admin", "password": "admin", "groups": ["sudo", "docker"]}]
}

"Create a Fedora desktop with AI tools"

{
  "name": "AI Workstation",
  "base_image": "fedora-43",
  "features": ["desktop", "ssh", "python", "ollama", "claude-code"],
  "hardware": {"min_memory_gb": 16, "min_storage_gb": 64}
}

"Build a GxP-compliant medical device image"

{
  "name": "MedTech Station",
  "base_image": "debian-bookworm",
  "features": ["ssh", "desktop", "gxp", "hipaa", "audit-logging", "security-hardening"],
  "security": {"hardening_level": "strict"}
}