# Morph Cloud

Official pricing: https://cloud.morph.so/web/subscribe  
Category: agent-sandbox · Isolation: vm

## Pricing regimes (raw)

- **MorphVM instance / Devbox, pay-as-you-go MCU ($0.05)** (resource): $0.05/vCPU-h, $0/GiB-h
- **Morph Actions GitHub runner (fixed sizes, billed per runner-hour, no idle pause)** (sizes): Small 2 vCPU/4 GiB $0.15/h; Medium 4 vCPU/8 GiB $0.25/h; Large 8 vCPU/16 GiB $0.5/h [alt]

## Features

Yes: snapshots, memory snapshots, fork/clone, pause/resume, persistent disk, volumes, idle auto-stop, ≥24 h sessions, custom image (Docker or snapshot), start from your own snapshot, full VM (own kernel), Docker inside, nested virtualization, root, systemd, desktop GUI, SSH, HTTPS preview URLs, open internet, Windows, macOS, Python, Node.js, wake on request, live fork (no pause), memory fork, audit logs, scoped API keys, automatic snapshots, snapshots on demand, agent harness API, gVisor or VM (no shared kernel), inbound access rules, firewall inside (nftables), secret proxy, extra volumes, shared volumes

No: your own Docker/OCI image, browser, computer-use API, code interpreter, browser automation API, browser + desktop control, anti-bot stealth, CAPTCHA solving, residential IPs, preinstalled agents, public IPv4, raw TCP inbound, egress allowlist, static egress IP, self-hosting, BYOC, open source, GPU, arm64, GPU desktop, HTTP method/path egress rules, live resize, MCP server, hosted agent API (their own agent), secret proxy for any API

Unknown: everything else. Evidence (source + quote) per feature: https://battleships.dev/data/providers/morph-cloud.json → feature_evidence

## Caveats

- Plan names corrected vs older research: subscribe-page columns are Developer ($0) / Team ($40) / Scale ($250) (badges Free / Popular / Enterprise), confirmed by sandbox.watch and the EFS plan table.
- MCU = max(vCPU, ceil(RAM/4), ceil(disk/16)); encoded via ram_per_vcpu max 4 at $0.05/vCPU-h. Disk above 16 GiB per MCU raises the price (e.g. 4 vCPU with 80 GiB disk = 5 MCU) and is not captured; storage.included_disk_gib=16 is the 1-vCPU allowance (really 16 GiB per billed MCU).
- Plans carry non-spec pool_vcpu/pool_ram_gib/pool_disk_gib: org-wide resource ceilings (Developer 64 vCPU => max 16 concurrent 4-vCPU VMs). Devbox concurrency 8/32/128 applies to the Devbox product.
- Team/Scale are prepaid MCU bundles worth more than the fee (fee 40 -> $50, 250 -> $375 at list). Rollover/expiry undocumented; org is restricted when credits are exhausted unless topped up.
- Paused instances bill only the snapshot (RAM+disk size) at $0.00001/GB-h (~$0.0073/GB-month); stopped instances are released and free. Snapshots have no default TTL.
- Billing granularity, egress, IPv4, regions, compliance and enterprise pricing unpublished.
- Windows / macOS guests are 'experimental' in devbox docs with no published price; modes are linux-only.
- No free credit: Developer tier starts at 0 MCU.
- No benchmark entry in benchmarks.json; third party claims <250 ms snapshot restore.
- auto_stop_idle=true relies on user-set TTL (ttl_action=pause) + wake-on-request; docs recommend combining them 'to automatically pause idle instances' but do not say whether TTL resets on activity (it may be wall-clock). No built-in idle timer is documented (checked 2026-09-28).
- Disk re-checked 2026-09-28: "One MCU represents ... 1 vCPU-hour + 4 GB RAM-hours + 16 GB disk-hours" (cloud.morph.so/web/subscribe). Disk has no separate $/GB price: it is free up to 16 GiB per billed MCU (64 GiB on a 4-vCPU box) and beyond that raises the MCU count (1 MCU per extra 16 GiB = $0.003125/GiB-h). The engine cannot express a per-vCPU allowance, so disk_gib_month stays null; use the mcu_per_vcpu knob when disk binds.
- Egress re-checked 2026-09-28: not published on the subscribe page or docs (null).
- Devboxes (not bare instances) document a built-in auto-sleep: 'Devboxes can branch to hundreds of parallel replicas or scale to zero after a period of inactivity, preserving all application and memory state' and prototyping users can 'optionally disable auto-sleep / TTL policies'. The default idle period is not published. Bare MorphVM instances still need ttl_action=pause + wake-on for scale-to-zero.
- EFS estimates round non-zero sampled bytes up to the next GiB (min 1 GiB-month per filesystem once sampled); EFS billing applies only 'where EFS billing is enabled'.

## How this provider charges


Morph bills everything in **Morph Compute Units (MCU)**. The standard (pay-as-you-go) rate is **$0.05/MCU**.

A running instance or devbox burns MCU/h = **max(vCPU, ceil(RAM / 4 GiB), ceil(disk / 16 GiB))**. It is a max(), not a sum. So up to 4 GiB RAM and 16 GiB disk per vCPU come free, and whichever dimension is largest sets the price. A paused instance burns only snapshot storage (1 MCU = 5 TB-hours, so $0.00001/GB-h, or about $0.0073/GB-month).

Subscriptions are **discounted MCU bundles**. They are not seats.

**Plan-name correction.** The raw subscribe-page text reads `Free/Developer · Popular/Team · Enterprise/Scale`. Those are badge + name pairs for **3** columns, and each row has 3 values (0 / 1,000 / 7,500 MCU; $0 / $40 / $250). So the tiers are:
- **Developer** ($0)
- **Team** ($40)
- **Scale** ($250)

Three other sources agree: sandbox.watch, the EFS plan table (Developer/Team/Scale) and the billing-console limits. The older `providers/morph-cloud.json` read them as Free/Developer/Team/Enterprise. That has been corrected.

## Regime table

| # | Regime | When it applies | How billed | Numbers | Source |
|---|---|---|---|---|---|
| 1 | Pay-as-you-go MCU (any tier; the only option on Developer) | Every running MorphVM instance or devbox, beyond any included bundle | MCU/h = max(vCPU, ceil(RAM/4), ceil(disk/16)) × $0.05. Allocated, not utilisation. Granularity **unpublished** | $0.05/MCU. 4 vCPU/8 GiB/≤64 GiB disk = 4 MCU = **$0.20/h**. 4/8/80 GiB disk = 5 MCU = $0.25/h (disk-bound) | https://cloud.morph.so/web/subscribe |
| 2 | Developer tier | Default | $0/month. **0 starting MCUs**, so you must buy top-up credits before any compute runs. Org limits: up to 64 vCPU / 256 GB RAM / 1,024 GB storage. Devboxes: 8 concurrent / 32 total | $0 | /web/subscribe, https://sandbox.watch/p/morph |
| 3 | Team tier (prepaid bundle) | $40/month subscription | 1,000 MCU included ("20% discount, $50 value"), i.e. $0.04/MCU. Extra MCU pay-as-you-go at $0.05. Org limits 256 vCPU / 1,024 GB / 4,096 GB. Devboxes 32 / 128 | $40 → $50 of usage | /web/subscribe |
| 4 | Scale tier (prepaid bundle) | $250/month subscription (badge "Enterprise") | 7,500 MCU included ("33% discount, $375 value"), i.e. $0.0333/MCU. Overage $0.05. Org limits 1,024 vCPU / 4,096 GB / 16,384 GB. Devboxes 128 / 512 | $250 → $375 of usage | /web/subscribe |
| 5 | Custom / enterprise beyond Scale | Contact | **Not published** on the subscribe page. The API supports promo codes that grant a tier with an end date (`tier_grant_ends_at`) | null | API ref (llms-full.txt) |
| 6 | Paused instance (`pause`, `ttl_action=pause`) | Memory + disk preserved | "The VM stops consuming MCUs; you only get charged for the snapshot taken." Snapshot size = RAM + disk | $0.00001/GB-h (5 TB-h per MCU). Page example: 8 vCPU / 8 GB / 8 GB disk snapshot = $0.12/month | https://cloud.morph.so/docs/documentation/instances/pause-resume, /web/subscribe |
| 7 | Stopped instance (`stop`, `ttl_action=stop`) | One-off tasks | "No longer charged for the instance after it stops." Stop releases resources, so it is effectively deleted | $0 | https://cloud.morph.so/docs/documentation/instances/ttl |
| 8 | Retained snapshots (named, Infinibranch, build cache) | Any saved snapshot | Snapshot-hours at the same rate. Snapshot TTL (retention by last-used) auto-expires dormant ones. Default = no TTL (kept forever) | ≈ $0.0073/GB-month | /docs/documentation/instances/snapshot-ttl |
| 9 | Infinibranch branch / fork (`branch(count=N)`) | Clone a running VM N times | Each clone is a normal running instance billed at full MCU/h. Snapshot dedup across branches is not documented | N × MCU/h | /docs/documentation/instances/branch |
| 10 | Wake-on-HTTP / SSH (scale-to-zero) | Paused instance auto-resumes on request | Snapshot rate while asleep, full MCU while awake | as 1 + 6 | /docs/documentation/instances/wake-on |
| 11 | Resize via reboot (`instances.boot` with new vcpus/memory/disk) | Scale up for builds, down for serving | Billing switches to the new spec "immediately after the resize" | as 1 | /docs/examples/resizing-instances |
| 12 | GitHub Actions runners (Morph Actions) | CI jobs | Fixed sizes billed hourly from the same MCU pool | Small 2/4/40 = 3 MCU ($0.15/h). Medium 4/8/80 = 5 MCU ($0.25/h). Large 8/16/160 = 10 MCU ($0.50/h), all at $0.05 | /web/subscribe |
| 13 | Morph EFS (shared FUSE filesystem) | Named filesystems mounted across instances | Stored bytes, per plan, **billed separately from MCU**. The price *rises* with the plan | Developer $0.12, Team $0.15, Scale $0.18 per GiB-month. Filesystem quota 5 / 25 / 100 | https://cloud.morph.so/docs/efs/usage-and-billing |
| 14 | Credit exhaustion | Included + top-up MCU run out | Credits are prepaid (`included_remaining` + `topup_balance`). The org becomes `is_restricted` (access blocked) when billing is exhausted, unless top-ups are bought | n/a | API ref (UserCreditsResponse) |
| 15 | Windows / macOS (x86) guests | "Experimental support" in the devbox docs | No separate price. Presumably the same MCU formula (unverified) | null | /docs/devboxes/getting-started |
| 16 | Egress / ingress / IP | Any | **Not published**. HTTPS service URLs and SSH gateway included. No public IPv4 | null | — |
| 17 | Regions | — | No region selection documented. No multipliers | 1 | features/morph-cloud.json |
| 18 | Free credit | — | None. The Developer tier starts at 0 MCU (sandbox.watch: pay-as-you-go) | $0 | /web/subscribe |

## Gotchas

1. **max(), not sum.** RAM up to 4 GiB/vCPU and disk up to 16 GiB/vCPU are free.
   - A RAM-heavy box (e.g. 2 vCPU / 32 GiB) bills 8 MCU = $0.40/h, which is $0.0125/GiB-h.
   - A disk-heavy box bills $0.003125/GiB-h, which is about $2.28/GiB-month of *running* disk.
   - The Actions Medium runner (4/8/**80**) costs 5 MCU, not 4, because 80 GiB of disk exceeds 64.
2. **The plans are almost pure prepay.** Team gets $50 of MCU for $40, and Scale gets $375 for $250. Overage is at list price.
   - The only saving is $10/month (Team) or $125/month (Scale).
   - If you use less than the bundle, you overpay. Rollover/expiry of unused bundle MCU is **not documented**.
3. **The free tier has 0 MCU.** "Free" means you can sign up, not run compute.
4. **Plan limits are org-wide resource ceilings** (64 / 256 / 1,024 vCPU), not per-VM caps. 50 × 4 vCPU = 200 vCPU forces Team. Devbox concurrency (8 / 32 / 128) is a separate cap on the devbox product.
5. **Paused ≠ free, but nearly.** The snapshot includes **RAM + disk**. At $0.0073/GB-month, a 50-box fleet with 8 GiB RAM + 20 GiB disk each pauses for about $10/month. Stopping is free but loses state.
6. **Snapshots never expire by default.** Set a snapshot TTL for caches and checkpoints.
7. **EFS pricing is inverted.** Higher plans pay *more* per GiB ($0.12 → $0.18).
8. **30% CPU utilisation saves nothing.** Billing is on allocation.
9. Not published at all: billing granularity, egress, and any volume/enterprise pricing. The docs repeatedly say the billing console is the source of truth.
10. **Isolation is undocumented.** Docs say "MorphVM"; third parties say Firecracker. There are no compliance claims (SOC2 unverified).

## Worked example

Workload: 4 vCPU / 8 GiB, 50 concurrent × 8 h/day × 22 days = **8,800 instance-hours**, 30% util, 50 GiB snapshots retained, 100 GiB egress.

Assumptions:
- Disk ≤ 64 GiB, so 4 MCU/h and **35,200 MCU/month**.
- Instances are paused at night (554 h × 50). Each paused snapshot = 8 GiB RAM + 20 GiB disk (assumed) = 28 GB.
- Org vCPU need is 200, so Team (256 vCPU) or Scale is required. Developer (64 vCPU) cannot run 50 at once.

| Regime | Compute | Plan fee / bundle | Paused-state snapshots | Retained snapshots 50 GiB | Egress | **Monthly total** |
|---|---|---|---|---|---|---|
| Developer, pay-as-you-go | 35,200 × $0.05 = $1,760 | $0 | 50 × 28 × 554 × $0.00001 = $7.76 | $0.37 | unpublished | **not allowed** (64 vCPU org cap). $1,768.13 if it were |
| Team ($40, 1,000 MCU) | (35,200 − 1,000) × $0.05 = $1,710 | $40 | $7.76 | $0.37 | null | **$1,758.13** |
| Scale ($250, 7,500 MCU) | (35,200 − 7,500) × $0.05 = $1,385 | $250 | $7.76 | $0.37 | null | **$1,643.13** (cheapest) |
| Scale, stop instead of pause (state lost) | $1,385 | $250 | $0 | $0.37 | null | **$1,635.37** |
| Disk-bound variant (80 GiB disk → 5 MCU/h), Scale | (44,000 − 7,500) × $0.05 = $1,825 | $250 | 50 × 88 × 554 × $0.00001 = $24.38 | $0.37 | null | **$2,099.75** |
| Morph Actions Medium runner (alt product, 5 MCU/h) | same as the disk-bound row | | | | | ~$2,075 + snapshots |

Egress (100 GiB) is unpublished, so it is excluded (null). The 30% utilisation has no effect.

Sources: https://cloud.morph.so/web/subscribe · https://cloud.morph.so/docs/documentation/setup/plans · https://cloud.morph.so/docs/documentation/instances/pause-resume · https://cloud.morph.so/docs/documentation/instances/ttl · https://cloud.morph.so/docs/documentation/instances/snapshot-ttl · https://cloud.morph.so/docs/documentation/instances/branch · https://cloud.morph.so/docs/documentation/instances/wake-on · https://cloud.morph.so/docs/efs/usage-and-billing · https://cloud.morph.so/docs/llms-full.txt · https://sandbox.watch/p/morph