Optimal's MCP (Model Context Protocol) server lets Gemini securely pull your research data straight out of Optimal, so you can ask natural-language questions and dive deeper into your research.
These steps are written for Gemini. The same Optimal MCP server also works with other MCP-compatible AI assistants, including Cursor, ChatGPT, Microsoft Copilot, Claude, and other custom MCPs.
Before you start — important note
The consumer Gemini app (gemini.google.com) does not currently support adding custom MCP connectors. There's no personal "add a connector" option available today.
To connect Optimal to Gemini, you'll need to follow one of these two paths, depending on your setup:
Gemini Enterprise — for teams using Google's Enterprise AI platform. Set up by an admin as a "Custom MCP Server" data source.
Gemini CLI — for developers using Gemini from the command line.
Because Google is actively changing how Gemini handles custom tools, treat the steps below as a starting point and check Google's current documentation if something doesn't match what you see.
Option 1: Gemini Enterprise (admin setup)
This is done once by an admin, and then made available to your team.
In the Google Cloud console (or Gemini Enterprise admin area), go to Data Stores (also called Connectors) and select Add Data Store > Custom MCP Server.
Enter:
MCP Server URL:
https://app.optimalworkshop.com/mcpMCP Server Description: a short description (e.g., "Optimal Workshop research data")
Configure OAuth 2.0 authentication for the connection.
Select Create. Gemini Enterprise creates the data store — wait for its status to change to Active.
Go to Actions, click Reload custom actions, then select and Enable the tools you want available (all tools are disabled by default).
Add the data store to your Gemini Enterprise app. The first time a team member uses it, they'll be prompted to authorize with their own Optimal login.
Option 2: Gemini CLI (developer setup)
Open (or create) ~/.gemini/settings.json.
Add Optimal under mcpServers:
{
"mcpServers": {
"optimal": {
"httpUrl": "https://app.optimalworkshop.com/mcp"
}
}
}Restart Gemini CLI, then run /mcp to confirm Optimal's tools are listed and connect/authenticate when prompted.
Who can connect
Gemini Enterprise: an admin with Data Store / Discovery Engine permissions sets this up once for the team.
Gemini CLI: each developer configures their own local settings file.
What can Gemini read?
Gemini can only read your Optimal data. It can't create, edit, or delete anything in Optimal, and it only sees what your own user login already has access to.
Study setup & metadata — study configuration and metadata (names, methodology, dates, etc.); tags, observations, and notes added by your team
Participant results — task results, success/completion metrics, and clickmaps/paths for prototype & usability tests; aggregated survey response distributions for structured questions (radio, checkbox, rating, etc.). Open-text/free-form survey answers are not currently surfaced through the MCP.
Recordings & transcripts — session and transcript counts for interview studies. Raw recordings and playback (video/audio) are not retrieved directly — what's exposed is derived from them (counts, AI analysis), not the recording files themselves.
AI-generated outputs — AI-generated study summaries, plus categorized insights such as themes, pain points, opportunities, sentiment, behaviours, and mental models, each with source/citation counts; task-, section-, and study-level analysis
Example prompts
These are just a few examples to get you started. You're not limited to this phrasing or these topics. Ask Claude in your own words; it can handle a much wider range of questions across your Optimal data.
"What were the biggest usability issues across our last five prototype tests?"
"Summarize participant feedback for our onboarding flow."
"What themes emerged from our customer interviews this quarter?"
"Compare findings between our March and June studies."
Things to know
Gemini Enterprise's custom MCP support only works with the Streamable HTTP transport (which is what Optimal uses). The older SSE transport isn't supported.
Optimal's MCP and Google's custom MCP support are both evolving quickly, so steps may change. Reach out to your Optimal contact if something doesn't match what you see.
Questions? Please get in touch with our team or share your feedback via our feedback and features request form.