On this page
Plugins & MCP#
A Plugin is an integration attached to a Skill that gives your chatbot new abilities at chat time. Where a Knowledge Base lets the assistant answer from your content, a Plugin lets it fetch live data or call an external service mid-conversation. Plugins are available on Starter tier and above.
You install a Plugin from the Plugins page in the Console, then attach it to a Skill. The assistant gains the Plugin's tools for every widget that uses that Skill. Credentials you enter during setup are encrypted at rest using an account-scoped key — the raw values are never stored in plaintext.
Plugins run isolated from the core platform
Each Plugin runs in its own isolated environment. A failure in a Plugin degrades the current turn gracefully — it does not affect the rest of the assistant or other widgets on your account.
Google Sheets#
The Google Sheets Plugin connects your chatbot to a Google Spreadsheet. The assistant can read rows and columns from a sheet and include that data in its answers — useful for pricing tables, staff directories, event schedules, or any structured content you maintain in a sheet.
What you configure#
| Field | How to fill it in |
|---|---|
| Google account | Click Sign in with Google in the Console. Gydr requests read-only access to your spreadsheets. |
| Spreadsheet | Select a spreadsheet from the list, then choose which sheet tabs the assistant may read. |
Example#
A visitor asks “What are your current membership prices?” The assistant reads the pricing tab and answers with the up-to-date values — no re-deployment needed when your prices change in the sheet.
REST API#
The REST API Plugin lets you describe your own HTTP endpoints and have the assistant call them during a conversation. You define each endpoint — method, path, parameters, and authentication — and the assistant decides when to call it based on what the visitor is asking. This Plugin is available on Growth tier and above.
What you configure#
| Field | Notes |
|---|---|
| Base URL | The root of your API — HTTPS only, no private IP addresses. |
| Authentication | Choose one of: none, API key header, Bearer token, Basic auth, or OAuth 2.0 (client credentials or refresh token grant). |
| Endpoints | 1–10 endpoints per attachment. Each has a tool name, HTTP method, path (with {param} placeholders for path parameters), optional query parameters, and a description of what the response contains. |
Example#
Define an endpoint on your order-status API. The assistant calls it when a visitor asks about their order, passes the order number from the conversation as a path parameter, and replies with the live status.
Tool name: get_order_status
Method: GET
Path: /orders/{order_id}/status
Path params: order_id — "The order number the visitor provided"
Description: Returns the current fulfillment status and estimated delivery date.Shopify Storefront#
The Shopify Storefront Plugin turns your chatbot into a personal shopping assistant for your Shopify store. The assistant can search products, browse collections, manage a shopper's cart, and — when the shopper is signed in — show their order history. Checkout and payment always happen on Shopify's own page; the assistant only facilitates discovery and cart management.
What you configure#
| Field | Notes |
|---|---|
| Store domain | Your permanent Shopify domain (the *.myshopify.com one, not a custom domain). |
| Storefront access token | A public Storefront API access token from a Shopify custom app you create in your Shopify admin. Gydr never stores this unencrypted. |
| Cart write tools | Cart creation and modification tools are off by default. You opt in to each write tool individually in the Console once the Plugin is attached to a Skill. |
To show order history, your Shopify storefront page passes the shopper's Customer Account API token to the widget via Gydr.identify() at page load. Gydr holds this token only for the duration of that session and never stores it.
Example#
A shopper asks “Do you have the Merino wool sweater in size M?” The assistant searches your product catalog, returns matching variants with prices and availability, and — if cart write tools are enabled — offers to add the item to the shopper's cart. After the cart is updated, the assistant shares the checkout link to complete the purchase on your Shopify store.
MCP Connector#
The MCP Connector lets you connect your chatbot to any remote Model Context Protocol (MCP) server. Gydr connects to the server, discovers its tools, and lets you grant each tool individually to the assistant. The assistant then calls the granted tools at chat time, just like any other Plugin.
This is useful for connecting to any service that exposes an MCP server — internal tooling, third-party SaaS products, or your own custom server. The MCP Connector supports Streamable HTTP transport.
What you configure#
| Field | Notes |
|---|---|
| Server URL | The HTTPS URL from your provider's MCP documentation. Paste the address without a path and Gydr will try the conventional /mcp and /sse endpoints for you. |
| Authentication | Detected for you. Gydr asks the server how it authenticates and acts on the answer: for OAuth it registers itself with the provider and sends you to that provider's own sign-in page, so there is nothing to fill in. If the server instead wants a token you already hold, or is one of the few providers that requires an app you register yourself, the Console asks for that one thing — and only then. |
| Tool permissions | After connecting, Gydr discovers the server's tools and shows them in the Permissions tab. Read-only tools are enabled by default. Tools that can modify data are disabled by default — you enable each one explicitly. |
Example#
You have an internal support tool that exposes an MCP server. Enter its URL in the Console, click Connect — approving access at your provider if it asks — and Gydr discovers the available tools — for example, get_ticket_status and list_open_tickets. Enable those tools in the Permissions tab, attach the Plugin to a Skill, and your chatbot can now look up support ticket status in real time without any additional integration work.
The MCP Connector supports up to 30 tools per connected server. Tools are re-discovered each time you click Test connection & refresh tools in the Console — changes on the server do not appear automatically.