Unwrap.ai
by Unwrap.ai
Query customer feedback, surface top issues, chart trends and sample entries from your AI agent. 17 tools, OAuth sign-in with five read-only scopes, and two tools that create saved objects the vendor calls read-only.
Verified connector
Listed by Anthropic as a partner connector in its Connectors Directory.
Connection checked by Agentman on .
Anthropic states this reflects the level of review a connector received, not a security audit.
Connect Unwrap.ai via MCP
https://nlp.api.production.unwrap.ai/mcpWorks in any MCP-compatible client. In Agentman Studio it is one click — no config file to edit.
Unwrap.ai Tools & Capabilities (17)
Tool names from Anthropic's directory listing. This server requires sign-in, so we could not read tool descriptions or parameter schemas.
Frequently asked questions
Is the Unwrap.ai MCP connector really read only?
Not quite. Unwrap's documentation states all its tools are currently read-only, and every OAuth scope the server advertises ends in read. But the same doc table describes sample_feedback_entries as creating a named sample and classify_feedback_entries as creating sub-samples for each label. Those persist server-side, so the label understates what the connector does.
What does sample_feedback_entries actually create?
A named, saved sample of feedback entries drawn from a view. Unwrap documents two modes, a representative random sample and a targeted semantic sample. The sample persists because two other tools depend on it: analyze_entry_text runs an instruction across a saved sample, and classify_feedback_entries splits one into labelled sub-samples. Nothing leaves your workspace.
Does leave_feedback write into my Unwrap data?
No. Unwrap documents leave_feedback as sending feedback on the MCP directly to the Unwrap team. It writes outbound to the vendor, not into your feedback dataset, your views or your samples. It is the one tool on the list whose output an agent cannot read back, and the one whose destination is Unwrap rather than your own account.
What does an Unwrap view control when I connect the MCP?
A view is the access boundary. Unwrap documents list_views as returning the views the authenticated org has permission for, and every data tool takes a view_id. Your agent picks one view, then queries within it. Static org-wide OAuth credentials are the exception: Unwrap says those grant read access to all views in the org.
What OAuth scopes does the Unwrap MCP server request?
Five, and all five end in read. Its RFC 9728 descriptor advertises team teams read, groups read, charts read, filters read and entries read, retrieved live on 2026-08-18. No scope covers creating a sample, so the consent screen cannot express the difference between reading your feedback and saving a named subset of it.
Can I share one Unwrap MCP connection across my whole team?
Yes, through static OAuth credentials. Unwrap documents an admin-only flow that issues a client ID and secret from the Organizations page, giving every user of those credentials read access to all views in the org. That bypasses per-user view permissions, so an individual OAuth sign-in is the tighter option for most teams.
Why does my client not call the Unwrap tools automatically?
Ask for Unwrap by name in the prompt. Unwrap's docs suggest phrasing requests explicitly, such as asking Unwrap for your top issues, when a client does not pick the tools up on its own. Unwrap also recommends loading its guide prompt first, and marking read-only tools as always-allow so the client stops pausing for approval.
Sources
- Unwrap MCP documentation — https://docs.unwrap.ai/articles/6412771011-unwrap-mcp (fetched 2026-08-18) · retrieved 2026-08-18
- Live RFC 9728 protected-resource descriptor — https://nlp.api.production.unwrap.ai/.well-known/oauth-protected-resource (fetched 2026-08-18) · retrieved 2026-08-18
- Live RFC 8414 authorization-server metadata — https://nlp.api.production.unwrap.ai/.well-known/oauth-authorization-server (fetched 2026-08-18) · retrieved 2026-08-18
- Live unauthenticated probe of
https://nlp.api.production.unwrap.ai/mcpreturning 401 with aWWW-Authenticatechallenge — 2026-08-18 · retrieved 2026-08-18 - Anthropic connector directory — https://claude.ai/directory/92d00407-7fa7-49bc-8d90-55a59434099c (snapshot 2026-08-16) · retrieved 2026-08-16
Use in Agentman
Connect once and your agents call these tools on their own — on a schedule, in a workflow, with nobody at the keyboard.
Open in Agentman StudioServer Info
- Category
- Analytics
- Developer
- Unwrap.ai
- Tools
- 17
- Domain
- nlp.api.production.unwrap.ai
Using Claude Desktop or another MCP client? Setup docs — the connection URL above works anywhere.