Unthread

Search support conversations, monitor SLAs and read analytics.

Add to Agentman Studio

Opens Agent Studio, where connecting is one click. The connector URL below works in any MCP client.

Paste it into any MCP client. Setup docs

Resources

What the Unthread connector does

The Unthread MCP server connects Unthread — a Slack-based shared inbox that turns customer conversations into tracked support tickets — to Claude, ChatGPT and any MCP-compatible agent. Anthropic's directory lists six read tools covering conversations, analytics and custom fields. Unthread's own documentation names eight more, including three that write. Sign-in is OAuth.

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.

Unthread tools (14)

  • get_current_user
  • list_conversations
  • get_conversation_by_id
  • get_time_series_analytics
  • list_custom_fields
  • get_custom_field

Tool names from Anthropic's directory listing. This server requires sign-in, so we could not read tool descriptions or parameter schemas.

Limits

  • The two published tool lists are disjoint. Anthropic's six names and Unthread's eight share nothing. Neither list is complete, and we could not resolve them against a live handshake because the endpoint requires OAuth. Fourteen is the union of what is published, not a verified total.
  • The directory's Read Only label is contradicted by the vendor. Unthread documents three write tools. Do not size this connector's blast radius from the directory permission field.
  • One OAuth scope covers everything. Both scopes_supported lists advertise the single value mcp, so consent is all-or-nothing. Nothing at the grant separates reading a conversation from commenting on one, and no scope check can exist that we could observe. The only enforceable boundary is the Unthread role or the service account's project scoping.
  • We could not read tool schemas or safety annotations. No parameter-level detail for any MCP tool is published here. In particular, we cannot say whether the MCP comment tool exposes isPrivateNote at all, or what it defaults it to.
  • Three different Unthread MCP surfaces exist. The admin connector documented here, a customer-facing portal server at your own portal domain, and WebMCP, which drives the Unthread dashboard through a browser session. Only the first is the connector in Anthropic's directory. Instructions for one do not apply to the others.
  • Unthread publishes no rate limits for the MCP server or the REST API. We found none in the API reference. Absence of a documented limit is not a guarantee there is none.
  • Unthread's only documented MCP troubleshooting is one error. The docs cover an "Authorization failed" message and advise checking the signed-in account and pop-up blocking. Anything else is unsupported by vendor documentation.
  • llms-full.txt on the docs host is a false 200. It returns a 547KB SPA shell whose title is "Page not found", and appending .md to any docs URL returns the same near-identical shell. The machine-readable affordances that host advertises do not work; only llms.txt itself does.

Frequently asked questions

Can the Unthread MCP server send a message the customer sees?

Unthread documents a create_conversation_comment tool on the MCP server, and the REST endpoint it matches posts to the customer by default. That endpoint takes an optional isPrivateNote boolean, and Unthread states a private note posts to the Slack thread without sending a response back to the customer. Omitting the flag sends the message onward.

How many tools does the Unthread MCP server have?

Fourteen names across two lists that do not overlap. Anthropic's directory lists six read tools; Unthread's own documentation names eight different ones, including five survey readers and three writes. Neither list contains a name from the other. We could not run a live handshake because the endpoint requires OAuth, so no single verified total exists.

Is the Unthread connector read-only?

No. Anthropic's directory records the permission as Read Only and lists six read tools, but Unthread's documentation has a section headed what data is available to write naming update_conversation, create_conversation_comment and create_approval_request. The directory listing and the vendor's documentation disagree, and the vendor is describing the server it operates.

What OAuth permissions does the Unthread MCP server request?

One scope named mcp. Unthread's RFC 9728 resource descriptor and its authorization server metadata both advertise scopes_supported as the single value mcp, checked on 2026-08-22. That is one switch for the whole server, so nothing at the OAuth grant separates reading a conversation from commenting on one. Your Unthread role is the only real boundary.

Does the Unthread MCP server take a service or method as an argument?

No. Every tool name Unthread and Anthropic publish is a fixed operation against a named object, such as list_conversations or get_survey_analytics. None takes an endpoint, service or method string that would forward to the wider REST API. The tool count therefore means what it says, unlike connectors that wrap a whole API behind one dispatcher.

What do I need before connecting Unthread to Claude?

An Unthread account and the server URL. Unthread's setup steps are to open Settings, then Connectors, then Add custom connector, enter a name and the URL https://app.unthread.io/api/mcp, then sign in through the OAuth prompt. Unthread states the assistant can only access data you have permission to view, so your workspace role carries over.

Is the portal MCP the same server as the Unthread connector?

No, they are separate servers with separate URLs. The connector in Anthropic's directory is the admin server at app.unthread.io/api/mcp, used by your support team. The portal server runs at your own portal domain followed by /api/mcp and is for your customers, who can use it to read the knowledge base and file tickets.

Can the Unthread connector delete a conversation or a customer?

No published tool deletes anything. Neither Anthropic's six names nor Unthread's eight carries a delete, remove, merge or archive verb, even though Unthread's REST API documents DELETE as one of its four methods. Closing a conversation is different: it is a status change, which update_conversation most likely reaches, and it is reversible.

Why should I be careful pointing an agent at support tickets?

Because ticket text is written by your customers. Anyone who can open a conversation can put text in front of your agent, which makes retrieved conversation content untrusted input. Pair that with a documented comment tool whose customer-visible setting is the default, and a well-crafted ticket becomes a route to a message going back out.

Sources

More Productivity connectors

  • 6sense

    Ask an AI agent about 6sense account insights, buying signals and keyword trends.

  • Adobe Workfront

    Find, create, update and delete Workfront work items from an AI agent.

Browse all MCP connectorsClaude and ChatGPT setup guideBrowse AI skills

Product names and logos belong to their owners. Agentman is not affiliated with a connector's maker unless the page says the connector is built and run by Agentman.

Use it in an agent

Put Unthread to work.

Connect Unthread once and your agents call these tools on their own: on a schedule, in a workflow, with nobody at the keyboard.

Open in Agentman Studio