Connect to Claude

claude.ai and Claude Desktop

claude.ai and Claude Desktop both take a custom connector. The steps are the same in either one.

1. Open Settings, then Connectors, then Add custom connector.
2. Paste the URL https://userken.com/mcp and add it.
3. Claude opens userken to ask for access. If you are not signed in you get the usual magic link login first, and land back on the same page afterwards.
4. The consent screen names the application asking, your workspace and your role in it, and says plainly that the connector will run tools as you and spend your workspace's credits. Approve it.
5. The userken tools appear in the tool list. Ask for something.

You stay connected until you disconnect. To end it, open /app/settings#connected and remove the entry under Connected apps. The connector stops working immediately.

Claude Code and other clients

Claude Code and other MCP clients that do not do the connector sign-in use a personal API key instead. Create one at /app/settings under API keys. The key is shown once, at creation, and only its hash is stored, so copy it then. Keys look like uk_ followed by 32 characters.

In Claude Code:

claude mcp add --transport http userken \
  https://userken.com/mcp \
  --header "Authorization: Bearer uk_your_key_here"

In any other client that takes a remote MCP server with headers:

{
  "mcpServers": {
    "userken": {
      "url": "https://userken.com/mcp",
      "headers": {
        "Authorization": "Bearer uk_your_key_here"
      }
    }
  }
}

A key carries your identity, so audiences, surveys and focus groups created through it belong to you and your workspace, exactly as if you had made them in the dashboard. Revoke a key from the same settings page; revocation takes effect at once. Personal API keys are part of the Pro plan and above.

The tool list

What you can do from Claude. Everything the dashboard does. Build an audience from filters or from a plain English description, preview it, build its personas, refresh, rename or delete it, and compare it with another. Run a survey of your own questions or one of the templates, watch the run, drill into an answer with a follow-up, re-theme the open answers and export the results. Hold a focus group: draft a discussion guide, open the room, ask it questions, put one to a single participant, summarise it, close it and export the transcript. Read your data sources, uploads and the published surveys behind the accuracy work. Check what a job will cost, what your workspace has left and where the credits went. Rows you create in Claude are the same rows the dashboard shows, and the same credits pay for both.

This list is generated from the running server when the page is rendered, so it is what the server actually exposes right now, not a copy that drifted. Tools that write take your identity from the connector or the API key.

Audiences and personas

Surveys

Focus groups

Catalogue

Sources and data

Reviews and analysis

Credits and accuracy

Account

Things to ask

Things worth asking once you are connected:

• Which audiences do I have, and how big is each one?
• Ask the Reuters audience whether they would pay two dollars more for an ad free tier, and show me the split by persona.
• Run a concept test for this feature description against the news category, then suggest three follow-up questions.
• Start a focus group with four personas from my audience and open with a question about onboarding.
• Build me an audience of Android users of financial news apps who complain about login, over the last twelve months.

Endpoint details

Most people never need this section. It is here for anyone writing a client, running userken themselves, or debugging a connection.

The MCP endpoint requires a bearer token. A call without one is answered with a 401 and a challenge pointing at the protected resource metadata, which is how a client discovers where to sign in.

Token lifetimes

Only hashes of these are stored. If you run your own instance, see Self-hosting and operations for the two environment variables this depends on.