SiteGPT for AI agents
The support platform your AI agent can run
Build, train, and manage AI customer-support chatbots straight from Claude Code, Cursor, Codex, OpenCLAW, or Hermes. Drive it over the CLI, MCP, or REST API. No dashboard, no signup to start.
works with
> Add a SiteGPT support bot to acme.com $ sitegpt onboarding start https://acme.com trained 142 pages, no login needed set persona, brand colors, lead form ✓ ready preview & claim → https://sitegpt.ai/onboard/ab12cdEvery surface an agent needs
However your agent works, SiteGPT answers
Shell, protocol, or plain HTTP. Pick the surface your runtime speaks. They all reach the same account.
Agent-first onboarding
An agent builds a real, trained chatbot from a URL with no login, then returns a preview-and-claim link.
auth.md →Command-line interface
npm install -g @sitegpt/cli. Self-documenting, JSON-everywhere, --wait so agents never poll.
Explore the CLI →MCP server
sitegpt mcp exposes SiteGPT over the Model Context Protocol. Connect any MCP-aware assistant.
Server card →Agent skill
An installable skill that teaches an agent the full build-and-manage-a-chatbot workflow.
View the skill →REST API
A complete, documented v2 API. Everything the dashboard does, an agent can do over HTTP.
OpenAPI spec →Discoverable by design
llms.txt, a .well-known agent manifest, and AI-friendly robots, so agents can find and use SiteGPT on their own.
llms.txt →Why we're the best for AI agents
Not a chatbot with an API bolted on
SiteGPT was built to be operated by agents. Every surface, default, and guardrail assumes a machine might be driving.
Try before signup
Agent-first onboarding means an agent ships a working chatbot with zero credentials, so the human only signs up to keep it.
Three ways in
CLI, MCP, and REST API. Whatever your agent speaks, SiteGPT answers. No proprietary glue required.
Structured for machines
Every command returns typed JSON with stable codes and actionable hints, so an agent can reason about results, not scrape text.
Safe to automate
Scoped tokens, --dry-run previews, and explicit confirmations keep autonomous runs from doing damage.
One instruction
Tell your agent. It does the rest.
> Stand up support for docs.acme.com and turn on lead capture and human handoff • created chatbot (no account) • trained docs.acme.com + help center • enabled lead form + escalation • tested 3 questions, all answered ✓ done, claim link readyQuestions, answered
- Can I build a SiteGPT chatbot from Claude Code, Cursor, or Codex?
- Yes. Any agent that can run a shell or call an API can drive SiteGPT. Install the CLI, call the REST API, or connect over MCP. A published skill teaches the full workflow, and onboarding needs no login to start.
- How does SiteGPT work with MCP?
- Run sitegpt mcp to expose SiteGPT as a Model Context Protocol server, or point an MCP client at the published server card. MCP-aware assistants can then create and manage chatbots directly.
- Do agents need an account or API key to get started?
- No. Agent-first onboarding lets an agent create and train a real chatbot from a URL with no credentials, then hand back a link for the human to preview and claim.
- How do I use SiteGPT with OpenCLAW or Hermes?
- Both support the open agent-skills standard and MCP, which SiteGPT already ships. Install the SiteGPT skill or add the MCP server (npx -y @sitegpt/cli mcp), and your agent can build and manage chatbots.
- What can an agent actually do?
- Everything in the dashboard: create chatbots, add and sync knowledge, set persona, instructions, and branding, manage conversations and leads, and read usage and billing, all programmatically.
Point your agent at SiteGPT
Install the CLI, connect over MCP, or call the API, and let your agent ship a support bot.