MCP server

UserSound exposes a Model Context Protocol server so AI assistants like Claude and ChatGPT can work with your survey data directly: list surveys, read transcripts, search responses across surveys, and generate sentiment reports. Everything is scoped to your organisation and authorised with OAuth 2.1 — the assistant only sees what your account can see.

Works with Claude Claude Code ChatGPT Codex Cursor VS Code Windsurf Any MCP client

Connect a client

The server speaks MCP's streamable HTTP transport at:

URL



https://usersound.com/mcp

Point any MCP client at that URL. Clients that support OAuth discovery and Dynamic Client Registration (Claude, ChatGPT, MCP Inspector, and most others) register themselves automatically and open a browser window for you to sign in and approve access — no manual API keys.

Example client configuration

JSON



{
  "mcpServers": {
    "usersound": {
      "type": "http",
      "url": "https://usersound.com/mcp"
    }
  }
}

Authorization & scopes

The MCP server is an OAuth 2.1 resource server. Clients authorise with an authorization-code + PKCE flow against UserSound's own authorization server (discovered via /.well-known/oauth-protected-resource/mcp), and can register themselves at /o/register/ (RFC 7591). Tokens are granted the scopes you approve on the consent screen:

Scope	Grants
`interviews:read`	All read tools: surveys, responses, transcripts, search, reports, members.
`interviews:write`	`create_report` — start generating a sentiment report.
`billing:read`	`get_billing_status` — your organisation's subscription status.

Tools

Tool	What it does
`list_surveys`	List your surveys with response counts; filter by status, name or creator.
`get_survey`	One survey's settings and metadata.
`get_survey_setup`	The interview template and its questions (contributor roles only).
`list_responses`	One survey's responses with filters plus lifecycle and quality aggregates.
`get_response_transcript`	The full conversation transcript, summary and sentiment for one response.
`search_responses`	Search responses across all your surveys by name, sentiment, status or date.
`get_dashboard_summary`	Organisation-wide response status counts and the latest response.
`list_reports` / `get_report`	Sentiment reports for a survey; `get_report` is also the poll tool.
`create_report`	Start a sentiment report from a natural-language query or hard filters.
`get_billing_status`	Your organisation's subscription snapshot.
`list_organisation_members`	Active and archived members, plus pending invites.

Reports generate asynchronously: create_report returns a pending report id immediately; the assistant polls get_report until the status is complete (or failed) and then reads the report markdown and sentiment aggregates.

Try asking

Once connected, ask your assistant things like:

Prompt



What did people with negative sentiment say in the last 30 days?

Prompt



Which of my surveys got the most completed responses this month?

Prompt



Create a sentiment report for my most recent survey and summarise the findings

Access control

Every tool call runs as the user who authorised the client: organisation members see their organisation's surveys, and role permissions (viewer, member, admin, owner) apply exactly as in the dashboard.
Tokens expire and can be revoked at any time from your account's authorised-applications page.
An assistant can never widen access: cross-organisation ids simply return not_found.