Unwrap MCP

Last updated: April 16, 2026

What is an MCP?

MCP stands for Model Context Protocol, and is an open-sourced standard developed by Anthropic to connect external data and tools to AI models.

The Unwrap MCP gives you access to the same tools that power the Unwrap Assistant, allowing you to get the same great Assistant experience natively in your AI model of choice.

Connecting to Claude Desktop/Web

  1. Go into Claude, go to Organization Settings, and click Connectors

    Screenshot 2026-03-29 at 9.42.22 PM.png
  2. Click Add, and add the following:

    1. Name: unwrap (or whatever else you want to call it)

    2. url: https://nlp.api.production.unwrap.ai/mcp

  1. Click Add

  2. Then, go to the Unwrap connector, and click "Connect" if there's a connect button. Follow the redirect link, and let it redirect you back to Claude, and you should be authenticated!

    1. If there is a Configure button instead, you already probably authenticated the unwrap MCP, and you're good to go!

Connecting to Claude Code

  1. In a terminal, run claude mcp add --transport http unwrap https://nlp.api.production.unwrap.ai/mcp

  2. Enter claude code with the claude command

  3. Use the /mcp command, and authenticate the unwrap mcp

Connecting to ChatGPT

  1. Enable developer mode: https://developers.openai.com/api/docs/guides/developer-mode

  2. Go to Settings > Apps > Create an App

Screenshot 2026-04-06 at 4.42.08 PM.png
  1. Set https://nlp.api.production.unwrap.ai/mcp as your url

Screenshot 2026-04-06 at 4.43.57 PM.png
  1. Click "I understand and want to continue" and Add the App.

  2. Then, when you click Connect on the App, it should go through an oauth flow, where you login with your permissions and connect.

Connecting to Cursor

  1. Go to Cursor > Settings > Cursor Settings > Tools & MCP

  2. Add a new MCP server

Screenshot 2026-04-06 at 9.48.19 AM.png
  1. In your mcp.json, add the below:

{
  "mcpServers": { // This should already be present
    ...
    "unwrap": { // ADD THIS
      "url": "https://nlp.api.production.unwrap.ai/mcp"
    }
    ...
  }
}
  1. Then, you should see a Connect button (might have to reset Cursor to see this). Click connect, and you should get authenticated to use Cursor

Static OAuth Connection

If you want one shared set of credentials to use for all users in your entire org, you may create static OAuth credentials to connect to the Unwrap MCP. This will give read access to all views in your org, to any user that uses the credentials. You need to be an Admin of your Unwrap org to create the credentials.

Generate an OAuth Client ID and Secret from your Organizations page in Unwrap. On the Organizations page, the OAuth Credentials section is in the bottom right. You'll need a callback url for this. This usually will be listed on the MCP clients doc page. You need this to be able to use the static credentials with a client.

Once you get the static OAuth credentials, follow the steps on each MCP clients docs to add the static credentials in the right spot.

Screenshot 2026-04-06 at 9.46.44 AM.png

Usage

Views

The first thing Claude will try to figure out is what View you're asking about. It will use the list_views tool to list the Views that you have permissions for. Select the View that you want it to use for the rest of it's queries.

Querying data

Now that you've selected a View, start querying data! Ask any questions you want about your customer feedback in natural language, and Claude will be able to use the Unwrap tools to explore your teams feedback, and give you a data backed answer.

Suggested Questions

  • What are my top issues for the last 90 days?

  • Can you filter down to just issues about X, and show me how they've trended over the last 6 months?

  • Can you find me customer feedback around feature X, and then write me a priority ranked PRD, with customer anecdotes and links?

Connection Issues

If you're experience connection problems, try going to your Claude settings, find the Unwrap MCP, disconnect it, and reconnect it. If the issue still persists, please reach out to our support team.

Tools

All Unwrap tools are currently read only.

Tool

Description

Dependencies

list_views

Lists the authenticated org’s available views, including view_id, view_name, and org_name.

None

get_started

Call this first when using Unwrap. Returns the core usage guide, including the data model and recommended workflows.

list_views to get a view_id to initialize

list_feedback_integrations

Lists the feedback integrations connected to a view and summarizes source coverage.

list_views if you need to discover the view_id first

search_segments

Semantically searches segment fields and values relevant to a question.

discover_filter_schema; also list_views if you need the view_id

list_segment_field_values

Lists the exact allowed values for a segment field so filters use the correct spelling.

Usually discover_filter_schema to identify the field; also list_views if you need the view_id

discover_filter_schema

Returns all available filter fields for a view so clients can discover valid filtering options up front.

list_views if you need to discover the view_id first

search_groups

Semantically searches taxonomy groups to find issues or themes relevant to a query. Supports optional filtering.

list_views if needed; when filtering, use discover_filter_schema and typically list_segment_field_values first

search_actionable_groups

Returns the most actionable groups for a question, with optional filtering and numeric segment sorting. Useful for top issues and prioritization workflows.

list_views if needed; when filtering or sorting by segment, use discover_filter_schema and the relevant segment/value lookup first

plot_chart

Generates charts or tables over time or by breakdown for trends, rankings, share comparisons, and other visual analyses.

list_views if needed; when filtering or selecting numeric segments, use discover_filter_schema -> search_segments, or list_segment_field_values first

sample_feedback_entries

Creates a named sample of feedback entries, either as a representative random sample or a targeted semantic sample.

list_views if needed; if sampling with filters, use discover_filter_schema and list_segment_field_values first

analyze_entry_text

Runs a custom instruction or question across all entries in a saved sample and returns prevalence statistics plus example results.

sample_feedback_entries; also list_views if needed

classify_feedback_entries

Classifies entries in a saved sample into predefined labels and creates sub-samples for each resulting label.

sample_feedback_entries; also list_views if needed