Head to head · Agent harnesses · October 2026 research run

boxes.dev vs Unstoppable

boxes.dev scores 53.2 (D) on agent readiness against Unstoppable's 38.4 (E), and leads in 4 of 7 scored categories. Unstoppable leads on reliability, maintenance & community and transparency & trust. Both do agent harnesses.

Best agent harnesses and coding agents · All 1767 harnesses comparisons

Which one, for what

boxes.dev D

Best for Developers who want Claude Code or Codex running on cloud copies of their dev environment, started from Slack, Linear, a schedule, an API call or another assistant.

Ahead on

  • Schema & documentation, 57 against 30
  • Agent ergonomics, 60 against 5
  • Security & auth, 69 against 41
  • Payments & pricing, 50 against 45

Also in its favour

  • A hosted endpoint, with nothing to install

Watch for

No status page, incident history or SLA found

Unstoppable E

Best for A developer who runs Claude Code, Codex and Cursor on existing plans and wants parallel worktrees, pipelines and phone check-ins in one app.

Ahead on

  • Reliability, 45 against 32
  • Maintenance & community, 58 against 48
  • Transparency & trust, 67 against 60

Also in its favour

  • No key needed to call it
  • Free to start without a card
  • Open source

Watch for

No API, MCP server, SDK or documented CLI for an outside agent. A Linux command-line tool is offered for download with no documentation, and MCP tools for cloud tasks appear only in changelog lines

Why the edge

Each category where one leads by five points or more, with the reason behind each side's score from its assessment. The numbers are in the table below; this is what they stand for.

Agent ergonomics boxes.dev +55

boxes.dev 60

The MCP tool count is unknown. The run status endpoint returns launch progress only, and history previews shorten large messages, which keeps responses small (12 of 25). Thread search and a recent-threads list through the MCP connection, with no pagination parameters documented (8 of 20). Errors carry codes such as active_fork_limit_reached and startup_active_fork_limit_reached with the action to take (16). Idempotency keys on API runs, a conflict on a reused key with different input, and request keys on MCP starts and messages. Tool annotations could not be read (16 of 20). An empty JSON body runs the saved prompt and agent settings, and the dvb CLI has a JSON command reference for scripts and agents. No SDK (8 of 15). 12+8+16+16+8 = 60.

Unstoppable 5

These lines measure an interface an agent calls, and Unstoppable has none, so context cost, output controls and retries score nothing. For the person, a usage limit shows choices to wait for the reset or continue on credits (2 of 20). Guest use with no account, a first-run page that imports Claude and Codex projects, and few required settings. No SDK (3 of 15). Total 5.

Security & auth boxes.dev +28

boxes.dev 69

OAuth for the MCP server, with protected-resource metadata naming mcp.boxes.dev and four scopes (threads:read, threads:write, events:subscribe, settings:write), a per-project grant and removal from settings. Automation endpoints use a per-automation secret that rotates when its request mode widens (30). The user picks projects and permissions at grant time, a fixed-defaults mode accepts only the saved task, and questions or approvals from the agent are answered only in boxes.dev, never through the MCP connection (15). Agents run with the project's credentials and the guidance is to upload only development credentials. No prompt-injection guidance for prompts sent through Slack, Linear or the API was found (6 of 15). Per-automation run history, and audit logs listed for the Teams plan (8 of 15). A valid security.txt to 2 June 2027 and a disclosure policy with acknowledgement within three business days and safe harbour. No bug bounty and no certification found (10 of 20). 30+15+6+8+10 = 69.

Unstoppable 41

Harness reading of the framework checklist, as for the other harnesses. 30 for what leaves the machine by default, 20 for approvals and sandboxing, 15 for prompt-injection posture, 15 for audit and 20 for the security programme. Activity sharing is Off for every activity type by default and secrets are removed before upload. The repository's HOSTED-SERVICES.md now lists every telemetry flow with a switch in Settings, Privacy, Usage data, and the privacy policy names those switches, but all of them start on, PostHog session replays are recorded by default, LaunchDarkly gets the account and connected-provider emails by default and can only be switched to an anonymous context, and updates download automatically by default (15 of 30, unchanged, judgement call). Ask and Accept edits modes with allow and deny rules by command, file write, MCP tool and network host, deny winning, and safety rules that block force-pushes and destructive deletes. Every kind of chat starts in Full access, which skips prompts and safety rules, and local agents have no documented sandbox. Cloud worktrees get their own sandbox (8 of 20, per the 27 October check). The in-app browser stops agents acting on private sign-in pages. No other guidance on untrusted content from web pages, MCP servers or connected tools (3 of 15). Cost and usage panels, a team feed with tool-use metadata, approval rules with their origin, and an Audit Trail experiment in Labs (8 of 15, per the 27 October check). A SECURITY.md in the client repository sends reports to GitHub's private vulnerability reporting and states that fixes go into the latest release, with no acknowledgement time (7, was 2). No security.txt on unstoppable.ai per the 27 October check, the parent's expired, and no SOC 2 or ISO 27001 found. Total 41.

Schema & documentation boxes.dev +27

boxes.dev 57

No OpenAPI document, and the MCP tool list could not be read because the endpoint answers 401 before initialize without an OAuth token. The automation API's request and response fields are documented in prose tables (5 of 25). llms.txt with every help page listed (10). The help pages say when to target a new devbox, an existing devbox or an existing thread, when to steer a busy agent and when to wake a devbox first (16). Field tables with enum values such as backend (codex or claude), delivery (steer or next) and four run statuses, and unknown fields are rejected. The MCP schemas themselves were not seen (9 of 15). curl examples and named errors such as 409 devbox_unavailable, 409 thread_unavailable and 503 devbox_wake_failed, with what each means (12). /v1 in the API path on the home page, a single dated product update of 27 October 2026 and no changelog (5 of 15). 5+10+16+9+12+5 = 57.

Unstoppable 30

Graded on the interface an outside agent can use, which is none. No API, OpenAPI file, MCP server or documented command (0). unstoppable.ai/llms.txt lists the product, help and policy pages in Markdown (10). Twenty-two dated help articles explain for a person what each part does and when to use it, such as Code and Work modes and approval modes (5 of 20). No typed inputs an agent could call (0). Step-by-step examples for people and named error states such as usage limit reached and update failed, with no error responses for an agent (3 of 15). Numbered versions and a public dated changelog, which lists only four releases from 20 to 23 October 2026 (12 of 15). Total 30.

Reliability Unstoppable +13

boxes.dev 32

Hosted reading, for the remote MCP server at api.boxes.dev/mcp and the automation run endpoints. No status page found. The site footer says "All systems normal" with no link, and the trust centre at trust.boxes.dev draws in a script our reader could not run (0). No readable incident history (5). No request rate limits found. Plans cap awake devboxes (3 on the trial, 5 to 25 on paid plans), and the API answers 409 active_fork_limit_reached at the cap (5 of 15). The automation API takes an Idempotency-Key and returns the original run on a retry with idempotentReplay: true, and the MCP connection retries a lost start or message with the same request key. No Retry-After or backoff guidance found (12 of 15). No SLA found (0). Neither the MCP connection nor the automation API is marked beta, and the product launched in June 2026 per its own blog (10). 0+5+5+12+0+10 = 32.

Unstoppable 45

Read as local software, because the product is the desktop app the owner runs, as for the other harnesses. Downloads for macOS Apple Silicon, macOS Intel, Windows and, since November, Linux x64 and arm64 (.deb for Ubuntu 24.04 LTS, with SHA-256 checksums), version 1.19.0, plus iOS and Android apps in the stores, with the supported platforms stated. No package registry (20). The desktop and mobile client source has been public under MIT at github.com/unstoppableweb/unstoppable-code since 29 October 2026, but its only workflow closes pull requests, so there is no public CI running tests (0). Issues are now taken on that repository, which our reader couldn't open, so no record of open crashes could be counted (0, unchecked). Version numbers and a dated changelog, but the changelog still stops at 1.15.0 while 1.17.0 and 1.19.0 were tagged on 3 and 5 November, and breaking changes are not marked as such (10). Version 1.19.0 (15). The hosted parts (accounts, AI credits, cloud agents, Remote Control) have no status page. status.unstoppabledomains.com covers the domains business only, per the 27 October check. Total 45.

Maintenance & community Unstoppable +10

boxes.dev 48

Closed service. CLI @boxes-dev/cli 1.0.1134 published on 17 November 2026 (30). Only one dated product update was found, on 27 October 2026, and earlier CLI versions' dates were not read, so three releases in 90 days are not established (0). A support address, an in-app chat and a blog, with no public tracker or changelog (8 of 25). A CLI but no SDK, and no entry in the official MCP registry was looked for (5 of 15). The CLI is published from GitHub Actions with npm trusted publishing and signatures (5 of 10). 30+0+8+5+5 = 48.

Unstoppable 58

Read for a closed product with a public source export. Desktop 1.19.0, tagged source-desktop/v1.19.0 on 5 November 2026 and offered on the download page (30). Desktop 1.12.0 to 1.15.0 from 20 to 23 October, source tags for 1.17.0 (3 November) and 1.19.0 (5 November), and iOS updates since 11 August 2026 (20). A public changelog, a support email, in-app feedback, a Discord server and, since 29 October, GitHub issues, none of which we could read for replies (6 of 15, held). No SDK and no MCP registry entry (0). The source carries a pnpm lock file and no CI beyond closing pull requests (2 of 10). Total 58.

Transparency & trust Unstoppable +7

boxes.dev 60

Closed service under the Common Paper Cloud Service Agreement 2.1, dated 20 October 2026, with the standard terms' machine-learning clause deleted and California law (15). The privacy policy of 21 October 2026 gives retention by data type (session replay about 30 days, Slack data deleted within 14 business days of disconnecting), and the backup clause matches the terms (22 of 30). Price changes take 30 days' notice under the terms. No deprecation policy for the API or MCP server found (5 of 20). Service providers are named (Convex, Cloudflare Workers, Expo, OpenAI, Amplitude, Sentry, Google and Apple), but the host that runs devboxes is unnamed and data may be processed in the US and elsewhere (12 of 20). 15+22+5+12 = 54.

Unstoppable 67

Desktop and mobile client source published under the MIT licence since 29 October 2026, while the hosted services the apps use (accounts, AI gateway, relay, cloud agents) stay closed under terms of 6 October 2026 that name Unstoppable Domains, Inc. and Delaware law (27, was 15, as for an open core with closed parts). The privacy policy (effective 8 October 2026) is written for this product. It states activity sharing levels, Off by default, secret removal and encryption, configurable transcript retention, and now the telemetry switches. It gives no retention periods and names no subprocessors, although HOSTED-SERVICES.md says the policy describes how long data is kept (16 of 30). No deprecation policy. The terms allow removing parts of the service without notice. The changelog dates one retirement (GPT 5.5 to GPT 6.1 Sol) (3 of 20). HOSTED-SERVICES.md discloses each telemetry flow (PostHog, LaunchDarkly, Crashlytics, a per-turn signal and a first-launch notice) with what it carries and a switch for each, except that LaunchDarkly can't be turned off and the mobile first-launch notice has no switch. No data locations (15 of 20, was 6). Total 61.

Score by category

CategoryWeightboxes.devUnstoppableEdge
Reliability16%3245Unstoppable +13
Performance10%pendingpendingpendingnot scored in this run
Schema & documentation13%5730boxes.dev +27
Agent ergonomics13%605boxes.dev +55
Security & auth14%6941boxes.dev +28
Payments & pricing10%5045boxes.dev +5
Task success10%pendingpendingpendingnot scored in this run
Maintenance & community7%4858Unstoppable +10
Transparency & trust7%6067Unstoppable +7
Negative events≤1500
TotalD 53.2/100E 38.4/100

Facts side by side

Factboxes.devUnstoppable
KindMCP serverAgent harness
VendorBoxes, Inc.Unstoppable Domains, Inc.
Hosted endpointhttps://api.boxes.dev/mcpno (local only)
TransportsStreamable HTTP, HTTP
AuthOAuth or keyNone
PricingFreemiumFreemium
x402nono
LicenceProprietary. The service runs under the Common Paper Cloud Service Agreement 2.1, and the @boxes-dev/cli package is marked UNLICENSEDMIT for the desktop and mobile client source (github.com/unstoppableweb/unstoppable-code, from 29 October 2026). Hosted services stay closed under Unstoppable Domains, Inc.'s Terms of Service
Read-only variant documentednono
llms.txtyesyes
Last release2026-11-172026-11-05
Terms last updated2026-10-202026-10-06
Privacy policy last updated2026-10-212026-10-08
Customer content may train modelsnot found in the textnot found in the text
Terms restrict automated accessnot found in the textnot found in the text
Terms restrict benchmarkingnot found in the textyes
Terms or service can change without noticenot found in the textnot found in the text
Arbitration or class-action waivernot found in the textyes

How the scores have moved

Verdicts

boxes.dev

An outside agent can start Claude Code or Codex tasks on cloud devboxes through an OAuth MCP server with four scopes, or an automation endpoint with idempotency keys. No status page, SLA or tool schema could be read, and every route starts with a person setting up a project in the app.

Unstoppable

Unstoppable runs several coding agents side by side in separate worktrees on the owner's Claude, ChatGPT or Cursor plans, and its desktop and mobile clients are MIT source with a documented telemetry list. It has no API, MCP server or documented command an outside agent can call, and every chat starts in Full access.

Before you call either

boxes.dev

  1. Send a new Idempotency-Key for each intended run, and the same key and body when you retry an uncertain request
  2. Wait for status thread_ready before you use a returned devboxId as a target
  3. If you get 409 active_fork_limit_reached, free an awake devbox or wait. The plan's awake-devbox cap is full
  4. Read the thread in boxes.dev for the agent's answer. The status URL does not return it
  5. Ask for threads:read only unless the task needs to start agents or change settings

Unstoppable

  1. Don't plan to drive Unstoppable from another agent. It has no documented API, MCP server or CLI. Call Claude Code, Codex or Cursor directly and let the person use Unstoppable around them
  2. Inside Unstoppable, check the approval mode before acting. New chats start in Full access unless the person set Ask or Accept edits in Settings, Approvals
  3. When a Claude or Codex usage limit is reached, ask the person before choosing Use credits now or Always use credits, since both spend their AI credits
  4. Keep each task in its own worktree and let the person review the diff and press Create PR, rather than pushing to a shared branch
  5. If the person cares what leaves the machine, point them to Settings, Privacy, Usage data, where each telemetry flow starts on

Questions

Which is better for AI agents, boxes.dev or Unstoppable?

boxes.dev scores 53.2 (D) on agent readiness against Unstoppable's 38.4 (E), and leads in 4 of 7 scored categories. Unstoppable leads on reliability, maintenance & community and transparency & trust.

Can an agent call boxes.dev and Unstoppable without installing anything?

boxes.dev has a hosted endpoint at https://api.boxes.dev/mcp. No hosted endpoint is listed for Unstoppable.

Are boxes.dev and Unstoppable open source?

No open-source release is listed for boxes.dev. Unstoppable is open source (MIT for the desktop and mobile client source (github.com/unstoppableweb/unstoppable-code, from 29 October 2026). Hosted services stay closed under Unstoppable Domains, Inc.'s Terms of Service).

Other comparisons with boxes.dev or Unstoppable

Machine-readable

For companies

Do agents find, use and choose your tools?

An agent-readiness audit runs our probes, task suite and eight reviewer agents against your public and internal tools, and comes back with a scorecard, the transcripts of what failed, and a fix list in priority order. From $2,500, re-run included. We never take payment to move a rank. We do help companies earn one.