The Agent ExaminerIndependent Review Authority
ManagedNot yet hands-on tested

MyClaw

One-click, 24/7 managed hosting that runs an open-source OpenClaw agent on a dedicated server for you — zero DevOps.

In brief

MyClaw is a managed agent platform: One-click, 24/7 managed hosting that runs an open-source OpenClaw agent on a dedicated server for you — zero DevOps. Hosting is managed cloud hosting on kubernetes — one statefulset plus a dedicated persistent volume per bot; a free tier is available to start; and it's unsupported on MCP. Best suited to teams that want the platform to host and operate the agent for them.

What it is

MyClaw (myclaw.ai) is a single-tenant managed-hosting wrapper that runs an open-source OpenClaw (or Hermes) agent on a dedicated, isolated per-user server — a managed Kubernetes StatefulSet with a persistent volume per bot. It packages install, config, updates, and backups behind a simplified web UI reachable over email, Telegram, WhatsApp, and browser, aimed at non-technical power users who want an OpenClaw agent without setting one up. It is explicitly independent of and unaffiliated with OpenClaw, and is bring-your-own-LLM-key: you pay for hosting, model tokens are billed to you separately by the provider.

At a glance

7/7 facts sourced
Pricing
Hosting only, LLM tokens billed separately by you. Annual plans (default "Save 45%"): Lite $199/yr ($16.6/mo, reg. $29/mo), Pro $399/yr ($33.3/mo, reg. $59/mo), Max $799/yr ($66.6/mo, reg. $119/mo), Ultra $1,599/yr ($133+/mo, reg. $239/mo). No free tier — 7-day money-back only. Because model tokens are uncapped and billed by the provider, real all-in cost for an always-on agent is higher than the sticker; other MyClaw-branded surfaces publish different figures.source ↗
Hosting model
Managed cloud hosting on Kubernetes — one StatefulSet plus a dedicated persistent volume per bot, one bot per instance. "Stop" scales to zero while retaining data; independent scaling, hot config reload, and log streaming are offered. No self-host path and no published SLA are documented; data is on a 30-day PVC retention after which it is unrecoverable.source ↗
Memory & state
OpenClaw's built-in file memory persisted on the per-bot persistent volume. No managed vector store or knowledge graph is offered.source ↗
MCP support
Nonesource ↗
Integrations
Around six documented messaging channels — Discord, Telegram, Slack, and WhatsApp/Messenger/Instagram (the Meta ones via browser automation rather than official APIs). No voice or phone. Marketing cites "100+ integrations," but the roughly 20 tool connectors shown are OpenClaw's own ClawHub connectors, not a proprietary framework; channel credentials are entered manually rather than provisioned via per-agent OAuth.source ↗
Model providers
Bring-your-own-key across five providers, all pass-through: Anthropic, OpenAI, Google, Groq, and OpenRouter. Keys are AES-256-GCM encrypted, never returned to the browser, and rotatable without restart. There is no proxy, metering, per-tenant policy, credit pool, or fallback routing — no pooled or bundled model access.source ↗
Open source
Nosource ↗
1.7of 5

Editorial scorecard

Overall 1.7 / 5

Our opinion — the same 0–5 rubric is applied to every platform, including Alfe. See the rubric.

Developer experience3/5
Pricing transparency & value2/5
Scalability2/5
Memory & state1/5
Integrations2/5
MCP nativeness0/5

Strengths

  • Genuinely frictionless setup — a working hosted agent in roughly 30–60 seconds with no Docker, Python, ports, or terminal.
  • Clean isolation primitive: one Kubernetes StatefulSet and a dedicated persistent volume per bot, with scale-to-zero "stop" that retains data, hot config reload, and log streaming.
  • Honest bring-your-own-key economics for users who already hold model keys — you are charged only for hosting, and keys are AES-256-GCM encrypted and never returned to the browser.
  • Stays on upstream, unforked OpenClaw, so users get the real runtime plus the ClawHub skills ecosystem; the operator also ships a genuine open-source content engine (openclaw-master-skills, openclaw-guardian).
  • Multi-client pairing (browser, mobile, and CLI over WebSocket on a personal subdomain).

Trade-offs

  • Reliability is the recurring complaint — reviews and a real paid-user report flag early stability issues, frequent server reboots, and support lag.
  • No pooled or bundled LLM — bring-your-own-key only, across just five pass-through providers, so uncapped model-token spend is billed separately and all-in cost is unpredictable.
  • Thin platform depth: no MCP or self-bootstrap, no managed memory service (only OpenClaw file memory), and no teams, projects, orgs, RBAC, or fleets — pricing is N bots on one personal account.
  • Few first-class channels (about six), no voice or phone, and the Meta channels rely on fragile browser automation rather than official APIs.
  • Corporate opacity and brand fragmentation — no named founders or legal entity surfaced, and several unrelated "MyClaw" vendors share the name with inconsistent pricing, making due diligence hard; despite "full code control" marketing, no hosting or orchestration code is open source.

Our take

Editorial opinion — not a hands-on test result

In our opinion, MyClaw is a narrow but competent thing: a single-tenant "run OpenClaw on a box for you" hosting wrapper aimed at non-technical power users, not a multi-agent platform. It does the packaging genuinely well — setup is frictionless, the per-bot Kubernetes isolation is a sound design, the bring-your-own-key economics are transparent for those who already hold model keys, and the operator's open-source tooling (openclaw-guardian, openclaw-master-skills) is a real community strength. The trade-offs, on our reading, are depth and trust: no MCP, no managed vector or knowledge-graph memory, no teams or fleets, only around six channels, bring-your-own-key spend that is billed separately and hard to predict, a recurring reliability complaint in reviews, and a fragmented, corporately opaque brand shared with several unrelated vendors. Held to the same rubric as everything else here, it lands low-to-mid: fair on developer experience for its scope, but behind the managed leaders on memory, integrations, MCP, and scalability.

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

Does MyClaw include the LLM, or do I bring my own key?
You bring your own key. MyClaw charges for hosting only — model tokens are billed to you separately by the provider (Anthropic, OpenAI, Google, Groq, or OpenRouter). There is no pooled or bundled model access and no credit pool.
Does MyClaw support MCP?
No. MyClaw does not offer native MCP or a self-bootstrap flow. It hosts an OpenClaw agent with OpenClaw's own file memory and ClawHub connectors.
Is MyClaw open source?
The hosted runtime is upstream open-source OpenClaw, but MyClaw's own hosting and orchestration control plane is closed — the LeoYeAI/MyClaw repository is a marketing/README-only project. "Full code control" refers to shell access to your own instance, not to the platform.

Compare MyClaw

Sources

The Agent Examiner editorial · last updated 2026-07-03. 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.