Run Customer Panels From Your AI Agent

This guide shows how to run synthetic customer panels directly from Claude, ChatGPT, or Cursor without leaving the agent surface. Setup takes about five minutes per agent. Once connected, your agent can create panels, ask questions, and return findings as part of any workflow.

The Minds MCP server exposes twelve tools covering everything from creating an AI persona to running a multi-segment panel and exporting results. Everything below assumes you have a Minds account. Sign up at getminds.ai if you don't.

What You Can Do Once Connected

Your agent gains the ability to:

• Create AI personas representing specific customer types
• Group personas into research panels
• Ask a panel any question in natural language
• Run multi-segment comparisons across personas
• Export results as Markdown, PDF, or JSON
• Pull historical panel analytics

The agent calls these tools in sequence. A typical flow: an agent receives a brief, creates a panel matching the target audience, runs three research questions, and returns a synthesized summary. Total time, end-to-end, is usually under two minutes.

Setup for Claude Desktop

Claude Desktop uses the mcp-remote bridge to call the Minds remote server with an API key.

Step 1. Generate a Minds API key in your account settings under API Keys. Keys start with minds_.

Step 2. Open the Claude Desktop config file. On macOS this is at ~/Library/Application Support/Claude/claude_desktop_config.json. On Windows it's at %APPDATA%\Claude\claude_desktop_config.json.

Step 3. Add the Minds MCP entry under mcpServers:

{
  "mcpServers": {
    "mindsai": {
      "command": "npx",
      "args": [
        "-y", "mcp-remote",
        "https://getminds.ai/mcp",
        "--header",
        "Authorization: Bearer minds_YOUR_API_KEY"
      ]
    }
  }
}

Step 4. Restart Claude Desktop. The Minds tools will appear in the tool picker. Test by asking Claude "list my Minds panels", it should call list_panels and return your existing panels.

Setup for Claude Web

Claude Web uses the remote URL connector with OAuth.

Step 1. In Claude Web, go to Settings → Connectors → Add Custom Connector.

Step 2. Enter the URL https://getminds.ai/mcp and click Connect.

Step 3. You'll be redirected to a Minds login flow. Sign in. The connector activates immediately after.

Step 4. Test by asking Claude to create a panel.

Setup for ChatGPT

ChatGPT uses Connectors with OAuth 2.1 and dynamic client registration.

Step 1. In ChatGPT, go to Settings → Connectors → Create new connector.

Step 2. Enter the URL https://getminds.ai/mcp. ChatGPT handles client registration automatically.

Step 3. Sign in with your Minds account when prompted.

Step 4. In any conversation, enable the Minds connector for that thread. Tools are now available.

Setup for Cursor and VS Code

Cursor and VS Code Copilot both speak MCP natively, with the same mcp-remote bridge as Claude Desktop.

Step 1. Generate a Minds API key.

Step 2. In Cursor, open MCP settings (Settings → MCP) and add a new server with:

• Name: mindsai
• Command: npx
• Args: -y mcp-remote https://getminds.ai/mcp --header "Authorization: Bearer minds_YOUR_API_KEY"

Step 3. Restart Cursor. The tools appear in the agent's tool picker.

VS Code Copilot follows the same pattern via its MCP extension settings.

Running Your First Panel

Once connected, the simplest workflow is:

Step 1. Ask the agent: "Create a panel of three personas representing mid-market B2B SaaS marketing managers in Germany."

The agent calls create_panel with the right parameters. It returns a panel ID and confirms creation.

Step 2. Ask: "Ask this panel whether they would pay 200 euros per month for a tool that automates customer interviews."

The agent calls ask_panel. The panel responds. The agent summarizes the responses.

Step 3. Ask: "Export this panel session as Markdown."

The agent calls export_panel. You get a clean Markdown file with the full session.

You can chain these in a single prompt: "Create a panel of US fintech early adopters, ask them about onboarding friction, and export the result." The agent will run the three tools in sequence.

Available Tools

The Minds MCP server exposes these tools to the agent:

Persona tools. list_minds for listing existing AI personas, create_mind for creating new ones from a description or upload, chat_with_mind for one-on-one conversations, get_mind_status for checking creation progress.

Group tools. create_group for combining personas into a segment, list_groups for browsing existing groups.

Panel tools. create_panel for setting up a research panel, ask_panel for asking the panel a question, list_panels for browsing previous studies, get_panel_status for checking readiness, get_panel_analytics for pulling structural data, export_panel for output in Markdown, PDF, or JSON.

The agent doesn't need to know these names. It reads the tool descriptions from the MCP handshake and chooses based on the task. You phrase your request naturally and the agent picks the right tool.

Patterns That Work Well

A few patterns we see succeed across teams:

Brief-to-summary in one prompt. Compose the entire research brief, the question, and the desired output format in a single prompt. The agent runs the full sequence.

Iterative messaging tests. Run a panel once with a baseline message, then ask the agent to test variants. The agent reuses the panel and just calls ask_panel with new wording.

Comparative panels. Create two panels (e.g. early adopters vs late majority) and ask the agent to run the same question across both. The agent compares responses automatically.

Daily checks. Schedule the agent to run a brief daily panel question on a specific topic. The agent posts the result to Slack via a connected Slack MCP.

Auth and Cost

API keys use Bearer auth. Keys are scoped to a single Minds account. Revoke and rotate from your account dashboard.

OAuth flows for ChatGPT and Claude Web use the same backend. Tokens expire and refresh automatically.

Pricing follows your Minds plan. Each tool call counts against the same quota as a manual panel run on the web app. Plans start at five euros per month for individuals; Enterprise plans for teams using high agent volumes start at fifteen thousand euros per year. See pricing for details.

Troubleshooting

If tools list but never fire, the most common cause is a client bug rather than a server issue. The Open WebUI integration in particular has known regressions in some versions, and on the Native function-calling path it requires Bearer auth (the upstream model provider cannot complete the OAuth handshake on Minds' behalf). Switch the connector to API key auth, or disable streaming on the model.

If OAuth fails on first connection, the most common cause is a stale connector entry. Delete and re-add. The Minds OAuth handler issues fresh tokens on each registration.

If the agent calls a tool but the result looks empty, check that you have at least one persona or panel already created in your Minds account. Some tools, like list_panels, return an empty array if you have none.

What's Next

Once your agent is connected, the next step is composing it with other MCP servers. The most common combination is Minds plus an analytics MCP (PostHog or GA4) and a notification MCP (Slack). This lets the agent run a panel, validate the finding against real product data, and report the result to a team channel automatically. The full list of MCP servers worth connecting in 2026 is in our companion post on the best MCP servers for marketing and research agents.

If you're new to the category, the definition of agentic market research gives the one-page overview of why agents change the workflow at all.

That's the full agentic research loop. It runs without a human in the middle, takes minutes instead of weeks, and costs a fraction of a traditional study. The setup above is the only thing standing between most marketing and product teams and that workflow.