You can build a chatbot trained on your own content and run it from Claude, without opening a dashboard. SiteGPT exposes a remote MCP server that MCP-compatible AI assistants connect to over OAuth. Add one connector URL (https://sitegpt.ai/mcp), approve access in your browser, and Claude can list your bots, pull leads, and update knowledge for you.
iShort answer
SiteGPT trains a customer service chatbot on your content (websites, docs, YouTube, help centers) and keeps it synced. The platform speaks MCP, the standard Claude already uses for connectors, so managing that bot becomes a conversation: "pull yesterday's leads", "add the new refund policy", "make the persona more concise". Setup takes minutes. Claude is the confirmed client, and SiteGPT is listed in Claude's connector directory. ChatGPT (Developer Mode) and Perplexity (Pro+) support remote MCP connectors, but the SiteGPT ChatGPT app has not been approved.
Four steps from connector URL to managing a live chatbot in plain language.
Build the bot in SiteGPT
Train it on your website, files, YouTube, help center, and 12+ other sources. Auto-sync keeps it fresh.
Add the connector in Claude
Click Connect on SiteGPT in Claude's connector directory, or paste https://sitegpt.ai/mcp as a custom connector. Nothing to install or host.
Approve in the browser
Sign in to SiteGPT over OAuth. No API keys are pasted into a chat.
Run it by asking
List bots, review leads, update knowledge, and edit personas from the conversation.
What changes when your chatbot speaks MCP?
The Model Context Protocol is an open standard for connecting AI applications to outside tools and data. The MCP specification defines servers that expose tools for an AI model to call, and clients inside the assistant that call them. SiteGPT runs the server side, so your assistant can operate the chatbot directly.
Your chatbot's day job stays the same: it answers customers on your website from your own content. What changes is how you operate it. Without MCP, every check-in, knowledge update, or lead review means logging into a dashboard. With MCP, each one becomes a single request to the assistant you already have open.
Same chatbot, same data. The difference is how much of your day the routine work takes.
Without MCP
- Log into the dashboard every time you want to check bots, leads, or conversations
- Click through screens to update knowledge, then repeat for every content change
- Export leads to a CSV, then reformat them by hand for your CRM or team
- Persona and settings edits live in forms, separate from where you think and write
- Every task is a context switch away from the work you were doing
With MCP
- Ask Claude "list my chatbots" and get status without leaving the conversation
- Say "add the new refund policy page to the bot" and the knowledge updates
- Say "pull yesterday's qualified leads" and have them summarized, qualified, and drafted into follow-ups in the same chat
- Edit the persona in plain language: "be more concise, stop offering discounts"
- Approve OAuth once, and after that operating the bot is just asking
In practice, an "MCP chatbot" is an ordinary chatbot on a platform that exposes its operations as tools your assistant can call. There is nothing for you to host or maintain. A custom MCP chatbot build means running your own server, OAuth, and tool schemas, and SiteGPT delivers that outcome as a managed product.
Being operable by other assistants also moves a chatbot up the autonomy scale. Where that leaves it relative to a full agent is covered in AI agent vs chatbot.
How do you connect Claude to SiteGPT?
Setup takes a few minutes. The detailed walkthrough with screenshots is in how to build and manage a chatbot from Claude.
- Build the bot in SiteGPT. Point it at your content: websites, sitemaps, files, YouTube, GitHub, cloud storage, and help centers (12+ sources). Auto-sync keeps trained content current, so answers do not go stale.
- Add the connector in Claude. Two paths reach the same remote MCP server. The faster one takes two clicks: open Claude's connector directory, search for SiteGPT, and click Connect (the listing is at claude.ai/directory/connectors/sitegpt). What that listing means and what the connector can do on day one is covered in SiteGPT in Claude's connector directory. The manual path is to add https://sitegpt.ai/mcp as a custom connector in Claude's settings. Either way, nothing gets installed locally.
- Approve over OAuth. Claude opens a browser window where you sign in to SiteGPT and approve the connection for all chatbots or only selected ones. No API keys touch the chat, and you can revoke access at any time.
- Ask for something. Claude works through three tools.
searchfinds the right operation in the SiteGPT API v2 catalog.execute_readruns read-only calls such as listing bots and leads.execute_writehandles creates, updates, and deletes, and it asks for confirmation first.
The connection is role-aware. The assistant inherits the connecting user's dashboard permissions and nothing more. If a read-only teammate connects Claude, their Claude cannot edit knowledge or personas.
What can you run from Claude?
You can list bots, review leads, update knowledge, and edit personas against your live workspace, in your own words:
- "List my chatbots." Every bot in the workspace with its status, without opening the dashboard.
- "Pull yesterday's qualified leads from the website bot." Claude fetches the lead captures and can summarize, qualify, or format them for a CRM handoff in the same conversation.
- "Add our new refund policy page to the bot's knowledge." The bot picks up fresh content without a retraining project.
- "Make the persona more concise and stop offering discounts." Persona instructions update directly from chat.
Follow-ups are where this saves the most time. "Pull yesterday's leads" can become "now draft a follow-up email for the two enterprise ones" without switching tools. For ten more prompts in this style, from unanswered questions to a Monday summary for Slack, see support analytics questions to ask Claude.
Which assistants besides Claude can connect?
Claude is the confirmed client, and the connection to https://sitegpt.ai/mcp is tested end to end. Because the connector follows the open MCP standard, any assistant that supports remote connectors can connect. ChatGPT (Developer Mode) and Perplexity (Pro+) support remote MCP connectors. The SiteGPT ChatGPT app has not been approved, so ChatGPT is not a confirmed client.
Coding agents point at the same remote MCP server, and there is a setup guide for each:
| Where you work | Setup guide |
|---|---|
| Claude Code, in the terminal | SiteGPT for Claude Code |
| Cursor, in the editor | SiteGPT for Cursor |
| Codex, in CI and the terminal | SiteGPT for Codex |
| OpenClaw, as a standing agent | SiteGPT for OpenClaw |
| Hermes, as a persistent agent | SiteGPT for Hermes |
Each one uses the same three-tool interface (search, execute_read, execute_write) and the same OAuth grant. The permission model does not change when you switch clients. What the protocol changes about support work in general is covered in MCP for customer support.
An MCP chatbot platform follows the open standard the assistant ecosystem is converging on, so it does not depend on one vendor's plugin program.
SiteGPT lists an assistant as confirmed once its connection test passes, and the MCP server docs carry the current list. For everything an agent can operate end to end, see SiteGPT for AI agents. The AI agents for customer service hub covers the broader shift this fits into.
Can you manage the chatbot from code instead?
Yes. MCP is the conversational control surface, and the same operations are available programmatically. All three surfaces drive the same underlying chatbot:
| Surface | Best for | Where to start |
|---|---|---|
| MCP server | Operating the bot from Claude in plain language | MCP server docs |
| Agent API v2 | Pipelines and product integrations: create bots, sync knowledge, fetch leads from code | API v2 getting started |
| CLI | Terminal and CI workflows, scripted knowledge updates | CLI onboarding |
Wire recurring jobs such as nightly knowledge syncs and lead exports to the Agent API v2 or the CLI. Keep the MCP connection for ad hoc work. When those jobs run from your own backend, the API integration guide covers token scopes, webhooks, and conversation context. API access is included on Growth ($79/mo billed yearly, $129 billed monthly) and above. For the terminal path from install to a provisioned bot, see provisioning a support chatbot from your terminal.
Is it safe to connect Claude to your chatbot?
Yes, within a deliberately narrow access model. Three controls limit what an assistant can do with production customer data:
- OAuth in the browser. Authorization happens against your SiteGPT account, so no shared secret transits the chat. Disconnecting the connector ends the assistant's access.
- Least-privilege, role-aware scopes. The assistant inherits the connecting user's dashboard permissions and nothing more. It can never do something the person who connected it could not do.
- Per-user grants. Each teammate authorizes their own connection under their own role, so access traces to a person instead of a shared workspace key.
SiteGPT's broader security practices, including SOC 2 Type II, are documented at sitegpt.ai/security. For HIPAA-covered deployments, a signed BAA is available on Enterprise only. In a covered workspace, connected apps and chat integrations are off, so check the HIPAA docs for how that affects assistant access before you connect one. sitegpt.ai/hipaa is the authoritative source for availability.
Frequently asked questions
Can Claude really manage a production chatbot? Yes. Once Claude is connected to https://sitegpt.ai/mcp over OAuth, it can list your bots, pull leads, update knowledge, and edit personas against your live workspace. The chatbot keeps answering customers on your website as usual, and Claude becomes the control panel.
Which assistants support the SiteGPT MCP server? Claude is the confirmed client. You can add SiteGPT from Claude's connector directory or paste the server URL as a custom connector. ChatGPT (Developer Mode) and Perplexity (Pro+) support remote MCP connectors, but the SiteGPT ChatGPT app has not been approved, so ChatGPT is not a confirmed client. An assistant is listed as confirmed only after the connection test passes. Coding agents connect to the same server, with per-client guides for Claude Code, Cursor, Codex, OpenClaw, and Hermes.
What is an MCP chatbot? An MCP chatbot is a chatbot you can operate through the Model Context Protocol, the open standard that lets AI assistants call external tools. The bot answers your customers on your website, and you manage bots, leads, knowledge, and personas by talking to an assistant instead of clicking through a dashboard.
What is the difference between an MCP server chatbot and an MCP client chatbot? An MCP server chatbot platform exposes its capabilities as tools that outside assistants call. SiteGPT works this way. An MCP client chatbot connects outward to other MCP servers to fetch data.
Do I need to build a custom MCP server for my chatbot? No. A custom build means hosting a server, implementing OAuth, defining tool schemas, and maintaining all of it as the spec evolves. SiteGPT ships that as a managed product: a hosted remote MCP server with browser OAuth and role-scoped permissions.
What permissions does Claude get over MCP? The assistant inherits the connecting user's dashboard permissions and nothing more. Scopes are least-privilege and role-aware, and disconnecting the connector ends access.
Is connecting Claude to my chatbot platform secure? Authorization runs over OAuth in the browser, so no API keys are pasted into a chat. Access is scoped to the connecting user's role and revocable at any time. Security practices, including SOC 2 Type II, are on the security page.
How much does this cost? SiteGPT Starter is $39/mo billed yearly ($59 billed monthly). Growth is $79/mo billed yearly ($129 monthly), and Scale is $259/mo billed yearly ($429 monthly). Annual billing saves up to 40%. Every plan starts with a 7-day free trial, and the trial requires a credit card. Growth and above include API access alongside the MCP server. Full details are on the pricing page.
Can I still use the dashboard? Yes. MCP adds a control surface and leaves the dashboard in place. The dashboard, the Agent API v2, the CLI, and the MCP server all operate the same chatbot.
Sources
- Model Context Protocol specification for the definition of MCP and its host, client, and server roles
- SiteGPT MCP server page for the server URL, OAuth approval, the three tools, and role-scoped permissions
- SiteGPT MCP server docs for the confirmed client list
- Claude connector directory listing for the SiteGPT directory entry
- SiteGPT pricing for plan rates, API access by plan, and the 7-day trial