Fly.io Machines
Founded 2017
Hardware-virtualized containers that launch instantly and run only when you need them.
What it is
Fly Machines run apps and code as fast-launching, hardware-virtualized micro-VMs in datacenters worldwide, managed via a REST API or the flyctl CLI, with subsecond start/stop. They are commonly used to host agents and MCP servers.
At a glance
6/7 facts sourced- Pricing
- Usage-based, billed per second while running (e.g. shared-cpu-1x/256MB roughly $2/mo; performance-1x/2GB roughly $32/mo). Volumes are $0.15/GB-month and egress varies by region. No standing free tier is advertised.source ↗
- Hosting model
- Hardware-virtualized micro-VMs on Fly-operated servers across 18 documented regions on six continents.source ↗
- Memory & state
- Durable state via Fly Volumes (persistent block storage) and snapshots; Machines can be persistent or stopped (stopped root filesystem is retained). Machine RAM itself is ephemeral.source ↗
- MCP support
- Nativesource ↗
- Integrations
- A Machines REST API and flyctl CLI, Fly Launch for containerized deploys, and a native fly mcp command surface with client wiring for major editors.source ↗
- Model providers
- Not applicable — pure compute/hosting; commonly used to host MCP servers and AI workloads.source ↗
- Open source
- NoUnverified
Editorial scorecard
Overall 3.7 / 5
Our opinion — the same 0–5 rubric is applied to every platform, including Alfe. See the rubric.
Strengths
- Subsecond start/stop micro-VMs with per-second billing.
- 18 self-operated regions across six continents.
- First-class MCP support with client wiring for major editors.
Trade-offs
- No advertised standing free tier.
- Persistence needs separately provisioned Volumes; Machine RAM is ephemeral.
- Egress pricing varies widely by region.
Our take
Editorial opinion — not a hands-on test resultIn our opinion, Fly Machines is a good foundation when you want low-level, globally distributed compute for hosting agents or MCP servers and are comfortable operating your own runtime. It is compute, not an agent framework, so you supply the agent logic. The native fly mcp tooling is a nice touch for teams standardizing on MCP.
Sponsored
Alfe is a managed agent-hosting platform. If you want an agent hosted and running without wiring infrastructure yourself, see Alfe → (or read its dossier).
FAQ
- How are Machines controlled programmatically?
- Via the REST Machines API or the flyctl CLI.
- Can Fly host MCP servers?
- Yes — fly mcp launch runs a remote MCP server over stdio or Streaming HTTP with bearer-token auth.
Compare Fly.io Machines
Sources
- Fly Machines docs · accessed 2026-07-01
- Fly.io pricing · accessed 2026-07-01
- Fly MCP docs · accessed 2026-07-01
- Fly.io about · accessed 2026-07-01
The Agent Examiner editorial · last updated 2026-07-01. Facts are drawn from each vendor’s own public materials; scores are our editorial opinion (methodology).
Product and company names are trademarks of their respective owners and are used here only to identify the products under discussion.