Google Search connector

The google-search MCP adapter calls Vertex AI Search (Discovery Engine) via searchLite. Use it in workflow mcp_call steps to retrieve indexed content from your own data stores and feed ranked results into LLM steps for RAG-style answers. Unlike web search, results come only from content you have indexed in your Vertex AI Search app.

Prerequisites

A Google Cloud project with Vertex AI Search (Discovery Engine) configured and an indexed search app
A Google Cloud API key authorized for Discovery Engine
A Google Workspace connection in AgentRuntime (OAuth via the Google account card on Connections) or a connection profile with search credentials
project_contributor access in your workspace

Connect in AgentRuntime

Connect Google

On Connections, use the Google account card and complete OAuth if your deployment binds search through Workspace. See Google Workspace setup and Connections.

Enable Search

On your Google connection, enable the Search service, or create a connection profile with your Vertex AI Search API key, project ID, and app ID.

Add google-search MCP instance

Go to MCP, click Add instance for google-search. Open Instance config, wire your connection, and set ProjectID, AppID, and APIKey in the instance config if not inherited from the connection.

Test in a workflow

Smoke-test with mcp_call in Workflow Studio. Test with a simple search query before production workflows.

What you can build

Support knowledge lookup — Customer question in webhook → search against your help-center index → LLM answer with citations from results.
Internal doc assistant — User prompt → search with user_pseudo_id for personalization → rank and summarize top hits for a chat response.
Compliance check — Policy change draft → search for related procedures → LLM gap analysis → human_task for legal review.
Product catalog Q&A — google-sheets SKU sync to search index → search by feature or part number → return matching product snippets to the caller.

Tools

Tool	Description
`search`	Run a query against Vertex AI Search using Discovery Engine searchLite.

Example

Retrieve relevant documentation before an LLM summarizes an answer:

{
  "id": "search-kb",
  "type": "mcp_call",
  "name": "Search knowledge base",
  "tool_name": "search",
  "tool_args": {
    "query": "{{input.user_question}}",
    "user_pseudo_id": "{{input.session_id}}"
  },
  "depends_on": [],
  "timeout_s": 30
}

Configuration

Beyond OAuth, the adapter requires Vertex AI Search identifiers:

Key	Required	Description
`APIKey`	Yes	Google Cloud API key authorized for Vertex AI Search
`ProjectID`	Yes	Google Cloud project ID that owns the search app
`AppID`	Yes	Vertex AI Search engine (app) ID used as the collection engine name
`APIBase`	No	Discovery Engine API base URL override

Troubleshooting

Issue	Fix
`403` API key not authorized	Enable Discovery Engine API and restrict the key to the correct project
Empty `results`	Confirm the search app index is built and the query matches indexed content
Wrong tenant or project	Verify `ProjectID` and `AppID` match the engine in Google Cloud Console
Pagination needed	Use `next_page_token` from the response in a follow-up call if your workflow supports paging

Get started

Platform

Workflows

Guides

Integration setup

Connectors

AI & chat

API

Google Search connector

Prerequisites

Connect in AgentRuntime

What you can build

Tools

Example

Configuration

Troubleshooting

​Prerequisites

​Connect in AgentRuntime

​What you can build

​Tools

​Example

​Configuration

​Troubleshooting

​Related

Prerequisites

Connect in AgentRuntime

What you can build

Tools

Example

Configuration

Troubleshooting

Related