SiteGPT

Send a reply to the visitor of one conversation, as a human agent of the team. The visitor sees it in the chat widget, or on WhatsApp for a WhatsApp conversation. If the team does not handle the conversation yet, the reply takes it over first: the chatbot stops answering it, and the visitor sees that a human agent joined. Send only text that the user wrote or approved. A resolved conversation refuses replies, and so does a conversation on Slack, Crisp, Messenger, or another channel with its own agent inbox (reply there). Renders the conversation with the new reply. Uses POST /api/v2/chatbots/{chatbotId}/conversations/{threadId}/replies.

reply_to_conversation

Send a visitor message to a SiteGPT chatbot and get its answer — the data lane behind the inline chatbot preview's built-in chat. The first call (no threadId) creates a conversation and returns its threadId; pass that threadId on follow-ups to continue the same conversation. Messages are capped at 20000 characters (the endpoint's limit). On chatbots in human-support mode the message is delivered but answer is null — a human replies in the site widget. Uses POST /api/v2/chatbots/{chatbotId}/messages and POST /api/v2/chatbots/{chatbotId}/conversations/{threadId}/messages.

send_chat_message

Open the SiteGPT home: switch between your chatbots and work with conversations, escalations, leads, analytics, and knowledge sources in one place. Takes no arguments. Opens from the ChatGPT sidebar or on request. Uses GET /api/v2/chatbots (every page, up to 1,000 chatbots).

open_sitegpt

Run read-only SiteGPT API v2 calls (GET only) using JavaScript code. Targets the SiteGPT API v2 (spec: https://sitegpt.ai/api/v2/openapi.json, docs: https://sitegpt.ai/docs/api-reference/v2/getting-started). First use 'search' to find the right endpoints. Requests run through an authenticated host-side bridge scoped to your OAuth grant. Write methods (POST, PUT, PATCH, DELETE) are rejected here — use the execute_write tool for those. Available in your code: interface RequestOptions { method: "GET"; path: string; query?: Record<string, string | number | boolean | undefined>; body?: unknown; contentType?: string; rawBody?: boolean; } // OpenAPI 3.x spec with $refs resolved inline. // The spec object follows the standard OpenAPI 3.x structure. interface OperationObject { summary?: string; description?: string; operationId?: string; tags?: string[]; parameters?: Array<{ name: string; in: "query" | "header" | "path" | "cookie"; required?: boolean; schema?: unknown; description?: string; }>; requestBody?: { required?: boolean; description?: string; content?: Record<string, { schema?: unknown }>; }; responses?: Record<string, { description?: string; content?: Record<string, { schema?: unknown }>; }>; security?: Array<Record<string, string[]>>; deprecated?: boolean; } interface PathItem { summary?: string; description?: string; get?: OperationObject; post?: OperationObject; put?: OperationObject; patch?: OperationObject; delete?: OperationObject; head?: OperationObject; options?: OperationObject; trace?: OperationObject; parameters?: OperationObject["parameters"]; } interface OpenApiSpec { openapi: string; info: { title: string; version: string; description?: string }; paths: Record<string, PathItem>; servers?: Array<{ url: string; description?: string }>; components?: Record<string, unknown>; tags?: Array<{ name: string; description?: string }>; } declare const codemode: { spec(): Promise<OpenApiSpec>; request(options: RequestOptions): Promise<unknown>; }; Your code must be an async arrow function that returns the result. Example: async () => { return await codemode.request({ method: "GET", path: "/your/endpoint" }); } SiteGPT API v2 is scoped by the bearer token used for this MCP connection. Use search first, then execute_read for lookups and execute_write for changes — /api/v2 paths only.

execute_read

Search the SiteGPT API v2 OpenAPI spec. Targets the SiteGPT API v2 (spec: https://sitegpt.ai/api/v2/openapi.json, docs: https://sitegpt.ai/docs/api-reference/v2/getting-started). codemode.spec() returns $refs resolved inline. This tool only reads the local spec catalog; it performs no API calls. Types: // OpenAPI 3.x spec with $refs resolved inline. // The spec object follows the standard OpenAPI 3.x structure. interface OperationObject { summary?: string; description?: string; operationId?: string; tags?: string[]; parameters?: Array<{ name: string; in: "query" | "header" | "path" | "cookie"; required?: boolean; schema?: unknown; description?: string; }>; requestBody?: { required?: boolean; description?: string; content?: Record<string, { schema?: unknown }>; }; responses?: Record<string, { description?: string; content?: Record<string, { schema?: unknown }>; }>; security?: Array<Record<string, string[]>>; deprecated?: boolean; } interface PathItem { summary?: string; description?: string; get?: OperationObject; post?: OperationObject; put?: OperationObject; patch?: OperationObject; delete?: OperationObject; head?: OperationObject; options?: OperationObject; trace?: OperationObject; parameters?: OperationObject["parameters"]; } interface OpenApiSpec { openapi: string; info: { title: string; version: string; description?: string }; paths: Record<string, PathItem>; servers?: Array<{ url: string; description?: string }>; components?: Record<string, unknown>; tags?: Array<{ name: string; description?: string }>; } declare const codemode: { spec(): Promise<OpenApiSpec>; }; Your code must be an async arrow function that returns the result. Examples: // List all paths async () => { const spec = await codemode.spec(); return Object.keys(spec.paths); } // Find endpoints by tag async () => { const spec = await codemode.spec(); const results = []; for (const [path, methods] of Object.entries(spec.paths)) { for (const [method, op] of Object.entries(methods)) { if (op.tags?.some(t => t.toLowerCase() === 'your_tag')) { results.push({ method: method.toUpperCase(), path, summary: op.summary }); } } } return results; }

search

Create, update or delete SiteGPT API v2 resources (POST, PUT, PATCH or DELETE only) using JavaScript code. Targets the SiteGPT API v2 (spec: https://sitegpt.ai/api/v2/openapi.json, docs: https://sitegpt.ai/docs/api-reference/v2/getting-started). First use 'search' to find the right endpoints and 'execute_read' for any lookups. Requests run through an authenticated host-side bridge scoped to your OAuth grant. GET is rejected here — use the execute_read tool to read data. GitHub knowledge-source connections cannot be created or updated through this tool because they carry an access token — direct users to the SiteGPT dashboard for GitHub setup; OAuth-based connectors (Notion, Google Drive, …) can be created here and finish authorization in the browser. Available in your code: interface RequestOptions { method: "POST" | "PUT" | "PATCH" | "DELETE"; path: string; query?: Record<string, string | number | boolean | undefined>; body?: unknown; contentType?: string; rawBody?: boolean; } // OpenAPI 3.x spec with $refs resolved inline. // The spec object follows the standard OpenAPI 3.x structure. interface OperationObject { summary?: string; description?: string; operationId?: string; tags?: string[]; parameters?: Array<{ name: string; in: "query" | "header" | "path" | "cookie"; required?: boolean; schema?: unknown; description?: string; }>; requestBody?: { required?: boolean; description?: string; content?: Record<string, { schema?: unknown }>; }; responses?: Record<string, { description?: string; content?: Record<string, { schema?: unknown }>; }>; security?: Array<Record<string, string[]>>; deprecated?: boolean; } interface PathItem { summary?: string; description?: string; get?: OperationObject; post?: OperationObject; put?: OperationObject; patch?: OperationObject; delete?: OperationObject; head?: OperationObject; options?: OperationObject; trace?: OperationObject; parameters?: OperationObject["parameters"]; } interface OpenApiSpec { openapi: string; info: { title: string; version: string; description?: string }; paths: Record<string, PathItem>; servers?: Array<{ url: string; description?: string }>; components?: Record<string, unknown>; tags?: Array<{ name: string; description?: string }>; } declare const codemode: { spec(): Promise<OpenApiSpec>; request(options: RequestOptions): Promise<unknown>; }; Your code must be an async arrow function that returns the result. Example: async () => { return await codemode.request({ method: "POST", path: "/your/endpoint", body: { name: "..." } }); } SiteGPT API v2 is scoped by the bearer token used for this MCP connection. Use search first, then execute_read for lookups and execute_write for changes — /api/v2 paths only.

execute_write

Render a live, interactive preview of a SiteGPT chatbot directly in the conversation (MCP Apps extension). Call it after creating or inspecting a chatbot, passing the chatbot id — an RFC-4122 UUID or a 15-20 digit numeric id — and optionally the chatbot title for the preview header. Returns { chatbotId, widgetUrl }; hosts that support MCP Apps render the actual chat widget from the ui://sitegpt/chatbot-preview resource, and any client can open widgetUrl in a browser instead. Performs no API calls and does not verify that the chatbot exists — use execute_read for lookups. Note: chatbots that restrict embedding via Allowed Domains (Settings > General) block the inline frame; the preview then switches to a built-in chat driven by the send_chat_message tool, and shows a fallback with the direct link when that fails too.

preview_chatbot

Open the SiteGPT support inbox as a panel in this conversation: switch between your chatbots and browse conversations, escalations, and leads. The conversation or lead you open becomes context for this chat, so you can ask for a summary or a draft reply. Takes no arguments. Uses GET /api/v2/chatbots.

open_sitegpt_inbox

Change SiteGPT settings from the ChatGPT plugin settings page. `set` maps property names from read_sitegpt_settings (for example leadEmails__<chatbotId>) to new values. Every value is validated before anything is saved. Changes take effect at once on the live chatbot widget and on lead email delivery. Returns every value after the save. Uses PATCH /api/v2/chatbots/{chatbotId}/settings/{section}.

update_sitegpt_settings

Update the chatbot widget appearance: title, welcome message, placeholder, tooltip, brand/launcher/link colors (#rrggbb), and launcher position. Only the provided fields change; the tool returns the full updated appearance. This changes the LIVE widget customers see. Uses PATCH /api/v2/chatbots/{chatbotId}/settings/appearance.

update_chatbot_appearance

Upload one file (base64-encoded, max 1500000 bytes decoded — about 1.5 MB; the MCP transport caps request bodies, larger files go through the dashboard) as chatbot training knowledge. Ingestion is queued asynchronously; once trained, the content shapes the answers the live chatbot gives visitors. Uses POST /api/v2/chatbots/{chatbotId}/knowledge/files.

upload_knowledge_file