How do I connect SiteGPT to OpenClaw?

OpenClaw supports skills and MCP. Install the SiteGPT agent skill (npx skills add sitegpt/agent-skills) or add the MCP server (npx -y @sitegpt/cli mcp), then ask OpenClaw to build a chatbot for your site.

Do I need a SiteGPT account?

No. Agent-first onboarding lets OpenClaw build and train a chatbot from a URL with no credentials, then return a claim link.

Is it safe to let an OpenClaw agent touch customer-facing systems?

With the right boundaries, yes. The SiteGPT connection enforces them structurally: the grant is scoped to chosen chatbots, reads are free, and every write asks. The chatbot talking to your customers never runs on the agent; it runs on SiteGPT. Your agent operates the platform, so the blast radius of an agent mistake is a rejected write, not a rogue reply to a customer.

Does SiteGPT work with OpenClaw skills?

Yes. Skills are how OpenClaw packages know-how, and SiteGPT ships an agent skill (npx skills add sitegpt/agent-skills) that teaches the full build-and-manage workflow. Pair it with the MCP connection: the skill knows the plays, MCP provides the hands.

What can OpenClaw do with SiteGPT?

Create chatbots, add and sync knowledge, set persona and branding, manage conversations and leads, and embed the widget, all programmatically.

How should I scope what my OpenClaw agent can touch?

Grant it only the chatbots it operates: the approval flow lets you select specific ones instead of the whole account. Reads are free under the grant; writes always confirm. If the agent’s job changes, revoke and re-approve rather than widening an old grant, so the permission trail stays honest.

SiteGPT × OpenClaw

Run SiteGPT support in OpenClaw

Point OpenClaw at SiteGPT over MCP or the CLI and your agent can build, train, and manage customer-support chatbots end to end.

All agent surfaces →

openclaw agent

> Stand up SiteGPT support for acme.com   created + trained chatbot (no account)  configured persona and lead capture  ✓ ready, claim link returned

Connect in two minutes

Connect SiteGPT to OpenClaw

OpenClaw supports skills and the Model Context Protocol. Add the SiteGPT skill or MCP server and OpenClaw can build and manage chatbots.

install the SiteGPT skill

npx skills add sitegpt/agent-skills

OpenClaw is an AI agent runtime. Learn more at OpenClaw.

What OpenClaw is

OpenClaw in one minute, and where SiteGPT fits

OpenClaw is an open-source autonomous agent framework you run yourself: on your own server or box, chatting where you chat, running tools, and continuing to work between your messages. You own the whole machine, which is both the appeal and the responsibility: setup, upkeep, and the safety posture are yours.

Two systems give an OpenClaw agent its capabilities: skills, which package know-how it loads when a task calls for it, and MCP, which gives it live access to external products. They compose: SiteGPT ships an agent skill (npx skills add sitegpt/agent-skills) that teaches the full support workflow, while the MCP connection does the actual reading and writing.

Customer support is a natural standing job for an always-on agent: watch the support bot’s conversations, close its knowledge gaps, keep it trained, and report weekly. The agent runs the routine; the SiteGPT chatbot faces your customers; you approve the judgment calls.

Because you operate the machine, verification is yours too: run the MCP connection from the agent host once by hand before wiring it into standing work, and keep the grant scoped to the chatbots the agent actually manages. An agent that can reach everything is a liability; one bounded to its job is a colleague. And if you want this standing-agent model without operating the machine at all, MissionControlHQ (from the same team as SiteGPT) runs a managed squad of persistent agents that speak the same skills and MCP surfaces.

Zero to operated

Give your agent a standing job

An always-on agent deserves standing work. Support is the cleanest standing job there is: bounded, recurring, and measurable.

Wire the connection

Connect SiteGPT over MCP (HTTP or stdio) and install the agent skill: the skill teaches the workflow, MCP provides the hands, OAuth bounds every call to your own role.

Stand up the front line

The agent builds and trains the chatbot from your website URL, sets the persona, and hands you the embed snippet in chat.

Give it the morning routine

Every day: check unanswered questions, draft fixes from your docs, queue approved sources for training, flag escalations that waited too long. You approve; it executes.

Get the weekly numbers where you chat

Volume, deflection, captured leads, delivered in the channel your agent already lives in. No dashboard sessions, no forgotten Fridays.

Input → outcome

Your agent types the left side.
Your customers see the right.

agent session

> Stand up support for acme.com   created chatbot (no account)  trained 142 pages + help center  set persona + brand colors  ✓ live, claim link ready

🤖

Acme Support

Online · answers instantly

Do you ship to Canada?

Yes! We ship to Canada on all plans. Standard delivery takes 5-7 business days and duties are included at checkout.

Source: acme.com/shipping

Anything else I can help with?

Ask anything…

Works in any agent that can read a URL: it discovers the no-signup onboarding flow on its own.

Why SiteGPT for OpenClaw

An always-on agent can own support

OpenClaw agents persist. Beyond building the chatbot, yours can operate it: sweep unresolved conversations, refresh knowledge when your site changes, and flag what needs a human.

Skills and MCP, both native

OpenClaw speaks the open agent-skills standard and MCP, and SiteGPT ships both: install the skill for the full playbook, or add the server and let the agent discover the API itself.

No signup to start

OpenClaw can spin up a real, trained chatbot from a URL with no SiteGPT account. You only sign up to keep it.

Steal these

Prompts to hand your agent

Everything below runs against your live account once connected. Reads run freely; anything that changes data asks first.

“Every morning, check the support bot for unanswered questions and draft fixes.”

A standing job: gaps found, answers drafted from your docs, queued for your approval.

“Build a support chatbot for my site and wire it into the weekly report.”

Created, trained, and folded into the agent’s recurring summary.

“Escalations older than a day: summarize each thread and ping me.”

The wait-list with context, delivered wherever your agent messages you.

“Compare this week’s deflection rate to last week and explain any drop.”

The trend with the likely causes, pulled from real conversations.

OpenClaw + SiteGPT, answered

How do I connect SiteGPT to OpenClaw?: OpenClaw supports skills and MCP. Install the SiteGPT agent skill (npx skills add sitegpt/agent-skills) or add the MCP server (npx -y @sitegpt/cli mcp), then ask OpenClaw to build a chatbot for your site.
Do I need a SiteGPT account?: No. Agent-first onboarding lets OpenClaw build and train a chatbot from a URL with no credentials, then return a claim link.
Is it safe to let an OpenClaw agent touch customer-facing systems?: With the right boundaries, yes. The SiteGPT connection enforces them structurally: the grant is scoped to chosen chatbots, reads are free, and every write asks. The chatbot talking to your customers never runs on the agent; it runs on SiteGPT. Your agent operates the platform, so the blast radius of an agent mistake is a rejected write, not a rogue reply to a customer.
Does SiteGPT work with OpenClaw skills?: Yes. Skills are how OpenClaw packages know-how, and SiteGPT ships an agent skill (npx skills add sitegpt/agent-skills) that teaches the full build-and-manage workflow. Pair it with the MCP connection: the skill knows the plays, MCP provides the hands.
What can OpenClaw do with SiteGPT?: Create chatbots, add and sync knowledge, set persona and branding, manage conversations and leads, and embed the widget, all programmatically.
How should I scope what my OpenClaw agent can touch?: Grant it only the chatbots it operates: the approval flow lets you select specific ones instead of the whole account. Reads are free under the grant; writes always confirm. If the agent’s job changes, revoke and re-approve rather than widening an old grant, so the permission trail stays honest.

The same platform trusted with real support traffic