SiteGPT × Claude
Run your customer support from Claude
SiteGPT is in the Claude connectors directory. One click, one browser approval, and Claude manages your support chatbots: conversations, training, escalations, leads. No API keys, no setup.
> What chatbots do I have? search("list chatbots") execute_read(GET /api/v2/chatbots) ✓ 46 chatbots, newest: Regenerative Cell (Jul 22) > Show this week’s unresolved conversations execute_read(GET /api/v2/conversations?status=open) ✓ 7 open, 2 waiting on a humanAgents read this page too
First principles
What are Claude connectors?
Connectors are how Claude reaches beyond the chat window. Each one plugs Claude into an external product over MCP, the Model Context Protocol, and describes a set of tools Claude can use there: reading data, running searches, making changes. Claude discovers those tools from the connector itself, so there is nothing to configure and no prompt engineering to maintain.
Connectors are different from Claude apps and artifacts, which live inside the conversation. A connector reaches outward, to a system that keeps running when the chat ends: your CRM, your repo host, or in SiteGPT’s case, your customer support platform.
The directory is Anthropic’s catalog of reviewed connectors. Listing requires a review of the tool surface and permissions, which is why connecting from the directory needs no URL and no trust decision beyond the approval page. What a connector can access is decided there: you see the permissions, and the grant is bounded by your own role in the connected product. Claude never sees more of SiteGPT than you could see yourself.
The honest answer
Can Claude handle your customer support?
Mostly yes, and the honest part matters: Claude is a brilliant reasoner, but on its own it cannot sit on your website answering visitors at 3am. A chat session belongs to you, not to your customers. It does not know your product unless you feed it, and it forgets when the conversation ends.
The architecture that actually works splits the job in two:
- A front line that never sleeps. A support chatbot trained on your own content, embedded on your site, answering every visitor with sources. That is SiteGPT: it holds your knowledge permanently, retrains as your docs change, escalates to humans, and captures leads.
- Claude as the operator. With the SiteGPT connector, Claude becomes how you run that front line: reviewing what it answered, fixing what it missed, training it, and reporting on it, all in plain language.
Your customers get instant, grounded answers around the clock. You get to manage the whole operation from the assistant you already think in. That pairing, not a raw chat window, is what Claude customer support looks like in practice.
One click from Claude
Connect SiteGPT to Claude
No custom connector setup needed: SiteGPT is listed in the Claude connectors directory, reviewed and published by Anthropic’s process.
Open the directory listing
Go to claude.ai/directory/connectors/sitegpt or find SiteGPT under Settings → Connectors in Claude.
Connect and approve
Click Connect. The browser approval page shows exactly what is being granted: the permissions, and whether the grant covers all chatbots or only ones you pick.
Ask Claude
Try: "List my SiteGPT chatbots, then show this week’s unresolved conversations." The tools do the rest.
Prefer a custom connector? The endpoint is https://sitegpt.ai/mcp, and the MCP server page covers every client, including Claude Code and Claude Cowork.
Your support platform, conversational
What Claude does with SiteGPT connected
Your SiteGPT chatbot answers customers on the website. Claude, with the connector, becomes how you run it.
Triage in one question
Ask for the unresolved conversations, the ones waiting on a human, or every thread where the bot said "I don’t know". Claude pulls the live list with links.
Close knowledge gaps
When the support bot misses a question, Claude finds the gap, drafts the answer from your docs, and queues the source for retraining, with your approval on the write.
Reports on demand
Volume, deflection, escalations, captured leads: summarized from live data in whatever shape you ask for, instead of a dashboard session.
Changes in plain language
Spin up a chatbot for a new product, add training sources, adjust the persona, update custom responses. Claude composes the same audited API your dashboard uses.
Steal these
Five prompts to start with
Everything below runs against your live account the moment the connector is approved. Reads run freely; anything that changes data asks first.
“List my SiteGPT chatbots and flag any with failed training sources.”
A live inventory with the ones needing attention, from one read.
“Show me every conversation this week where the bot could not answer.”
The knowledge-gap list, clustered by topic, with links into the dashboard.
“Draft answers for those gaps from our docs and queue them for training.”
Claude drafts, you approve, the write goes through the same audited API.
“Summarize support volume, deflection, and captured leads for Monday standup.”
A report in your format, pulled live, no dashboard session.
“Create a chatbot for our new product site and train it on the docs.”
Setup by instruction: create, ingest, persona, ready to embed.
Safe by construction
The approval page is the contract
OAuth grant, not a pasted key
Connecting creates a revocable grant scoped to what the approval page showed. You never hand Claude an API key.
Reads flow, writes ask
Claude groups SiteGPT’s read-only tools under "allow always" while anything that changes data asks first. That split comes from the server’s own tool annotations.
A subset of you, never a superset
The grant cannot exceed your SiteGPT role, token scopes, or chatbot restrictions. Claude can do at most what you could.
How Claude groups the SiteGPT tools, from the server’s own annotations
SiteGPT connector permissions
- Reader · search, execute_read
- Read-only: list chatbots, pull conversations
- Writer · execute_write
- Creates, updates, deletes: always confirmed
Claude + SiteGPT, answered
- What are Claude connectors?
- Connectors are Claude’s way of plugging into external tools over the Model Context Protocol (MCP). A connected app gives Claude real capabilities: reading your data, running searches, making changes you approve. SiteGPT’s connector lets Claude manage your AI customer support platform: chatbots, conversations, training sources, and leads.
- How do I connect SiteGPT to Claude?
- SiteGPT is listed in the Claude connectors directory. Open claude.ai/directory/connectors/sitegpt (or find SiteGPT under Settings, then Connectors), click Connect, and approve access on the browser page. No URL pasting, no API key. Adding https://sitegpt.ai/mcp as a custom connector also works.
- What data can Claude access through the SiteGPT connector?
- Exactly what you approve, and never more than your own SiteGPT role allows. The approval page shows the permissions and whether the grant covers all chatbots or only ones you select. Reads (listing chatbots, pulling conversations) are marked read-only; anything that changes data sits in a separate tool that asks before running. You can revoke the grant at any time.
- What can Claude actually do with SiteGPT connected?
- Operate the platform in plain language: list your chatbots, pull this week’s unresolved conversations, find questions the bot could not answer, queue new docs for training, draft and update custom responses, check captured leads, and summarize support activity. It composes the same API your dashboard uses.
- Does the SiteGPT connector work in Claude Cowork?
- Yes. Connectors attach to your Claude account, so a SiteGPT connection made once is available to Cowork sessions too. See the Claude Cowork guide for the workspace-side walkthrough.
- Is the SiteGPT connector free?
- The connector itself adds no cost on either side: it is included with SiteGPT and free to connect in Claude. SiteGPT has its own plans, and you can start without an account at all: an agent can build a preview chatbot from your website URL first.
- How do I disconnect or revoke access?
- Remove the connector in Claude’s settings, or revoke the grant from SiteGPT. Either side ends Claude’s access immediately; your chatbots and data are untouched.
- Does it work in Claude Code as well?
- Yes, over the same server: run claude mcp add --transport http sitegpt https://sitegpt.ai/mcp in the terminal, or use the stdio variant via the SiteGPT CLI. The Claude Code guide covers both.
One click away from Claude
Connect from the directory and ask Claude to run your support.