SiteGPT · AI Agents
SiteGPT is agent-ready infrastructure
Point any MCP-compatible AI assistant at SiteGPT and it builds, trains, and runs an AI agent for customer service, trained on your content and live over OAuth. It resolves questions 24/7, captures leads, and escalates to a human when it matters.
7-day free trial on every plan. Claude connects today; ChatGPT and Perplexity support remote MCP connectors too.
See what your agent ships
One loop, end to end: SiteGPT trains on your content, answers a real customer question with a cited source, and offers a human handoff.
Trained on your content
- Website
- Help center
- YouTube
- Notion
- PDFs
- GitHub
OnlineTrusted by these leading companies
“We’ve got the bot dialled in. We’re using GPT-4, have an avenue for escalations to Zendesk, and so far I have no complaints.”
What an AI assistant gets when it connects to SiteGPT.
Remote MCP server
MCP-compatible assistants connect over OAuth and operate SiteGPT through tool calls.
Agent API v2
A machine-readable agent manifest, OpenAPI discovery, and an OAuth device flow.
A CLI built for AI agents
Scriptable commands an agent can run to manage chatbots, content, and leads.
Agent-first onboarding
An agent goes from a blank account to a live, trained chatbot without a human clicking through a dashboard.
Live in three steps
Connect
Point your MCP-compatible assistant at SiteGPT's remote server and authorize over OAuth.
Train
The agent adds your website, help center, and docs. Auto-sync keeps them current.
Operate
Your chatbot goes live, resolving questions, capturing leads, and escalating to a human on request.
What agent-ready means
Most chatbot platforms assume a human is driving: dashboards to click through, docs written for people, API keys to copy around. SiteGPT treats the AI assistant as a first-class operator. Four building blocks make that work.
Building block 1
Remote MCP server
SiteGPT exposes a remote MCP server that MCP-compatible AI assistants connect to over OAuth. Operate SiteGPT from Claude today: list bots, review leads, update knowledge, edit personas.
Because the connection uses OAuth, you grant and revoke access like any other authorized app. No long-lived secret sits in a conversation transcript.
Building block 2
Agent API v2
Agent API v2 publishes an agent manifest and OpenAPI discovery, so an agent can enumerate every endpoint and its schema without reading prose documentation.
The OAuth device flow issues scoped credentials to headless environments, which keeps automation out of the paste-an-API-key-into-a-chat pattern entirely.
Building block 3
A CLI built for AI agents
Coding agents drive the SiteGPT CLI as easily as people do. Its scriptable, composable commands let an agent in a terminal create chatbots, add training content, and pull leads as part of a larger workflow.
Building block 4
Agent-first onboarding
Setup does not assume a person is watching a browser. An agent can take a new account from zero to a trained, embedded chatbot: create the bot, point it at content, set the persona, and ship the embed snippet.
The same job, with and without an assistant as the operator.
Typical chatbot platform
- A human clicks through dashboards for every change
- You copy API keys into chats and scripts
- Integration docs target people, not agents
- Onboarding stalls without a person watching a browser
Agent-ready SiteGPT
- An assistant operates bots through MCP tool calls
- You grant access over OAuth and revoke it like any app
- A manifest and OpenAPI discovery describe every endpoint
- An agent onboards the account end to end
Where your agent runs
One remote MCP server serves every client, so the permission model does not change when you switch tools. Each of these has its own setup guide.
New to the protocol? MCP for customer support explains what changes when an assistant operates the platform instead of advising on it, and SiteGPT for AI agents is the product page for the whole surface.
AI agents for customer service, trained on your content
The building blocks are the means. The payoff is what the agent stands up: an AI agent for customer service, trained on your business content, that resolves questions around the clock instead of routing them to a queue.
That chatbot resolves customer questions from the content it is trained on, captures and qualifies leads, books time with your team, and escalates to a human with full context when a customer asks for one. Auto-sync keeps the training content fresh as your product changes, so answers stay current without anyone retraining by hand.
12+
Content sources to train on
95+
Languages supported
24/7
Coverage once the agent ships it
OAuth
Auth model, no pasted API keys
4
Agent-ready building blocks
7 days
Free trial on every plan
Works on the channels your customers already use
The same content-trained agent answers across your site and the chat tools your team already runs, so customer service is consistent wherever people reach you.
- Website widget
- Slack
- Zendesk
- Crisp
- Messenger
- Google Chat
- Freshchat
- Zoho SalesIQ
Trust and compliance
Audited controls for security and confidentiality.
GDPR-ready data handling for EU customers.
BAA available on the Enterprise plan.
Go deeper
Start with the hands-on walkthrough, then the concepts behind it.
Pricing
Every plan includes a 7-day free trial. Switch to annual billing to save up to 40%.
Starter
Try it on one bot
$59$39/mo
billed $468/yr
- 1 chatbot
- 4,000 messages/mo
- 1,000 pages
- 1 seat
Growth
Where agents plug in
$129$79/mo
billed $948/yr
- 2 chatbots
- 10,000 messages/mo
- 10,000 pages
- 4 seats
- API access and integrations
Scale
Higher volume and automation
$429$259/mo
billed $3,108/yr
- 3 chatbots
- 40,000 messages/mo
- 50,000 pages
- 10 seats
- Webhooks and daily auto-scan
Enterprise
Regulated and large teams
Custom
- Up to 10,000 chatbots
- Custom message volume
- 500,000 pages
- HIPAA BAA and priority support
Add-ons: remove SiteGPT branding for $39/mo, or add 5,000 messages for $39/mo. See the full pricing page for details.
Frequently asked questions
What does agent-ready mean?
Agent-ready means an AI assistant can operate the product directly instead of just describing it. SiteGPT exposes a remote MCP server, an Agent API with a machine-readable manifest and OpenAPI discovery, and a CLI built for AI agents, so an assistant can build, train, and run a production chatbot end to end.
Which AI assistants can operate SiteGPT today?
Claude connects to SiteGPT's remote MCP server today and can list bots, review leads, update knowledge, and edit personas. ChatGPT (Developer Mode) and Perplexity (Pro+) support remote MCP connectors as of 2026. Coding agents point at the same server, and SiteGPT publishes setup guides for Claude Code, Cursor, Codex, OpenClaw, and Hermes. SiteGPT is built on the open MCP standard, so any assistant that supports remote connectors can connect.
How does an AI agent authenticate with SiteGPT?
Over OAuth. An assistant connecting to the remote MCP server signs in through a standard OAuth flow, and the CLI and Agent API support an OAuth device flow for headless environments. You never paste an API key into a chat.
What can an agent do through the MCP server?
It can operate a SiteGPT account: list chatbots, review and qualify captured leads, update the knowledge the chatbot answers from, and edit chatbot personas. Combined with the Agent API and CLI, an agent can take a chatbot from creation to production.
Is SiteGPT ready for production and regulated environments?
SiteGPT is SOC 2 compliant and GDPR ready, and offers a HIPAA BAA on the Enterprise plan. See the SiteGPT security page for the details.
Point your agent at SiteGPT
Start a free 7-day trial, connect your assistant to the remote MCP server, and let it stand up a chatbot trained on your content.