Search your conversations and read their labels from your MCP client

Until now, the MCP server could fetch one conversation if you already had its id. You can now ask your own Claude, or any MCP client, to find conversations and read what the labels say about them.

Six read-only tools.

• Search conversations across all your sites, or one. Filter by dates, channel, agent, language, shopper rating, title keywords, labels and label scores. Results come newest first, with a link to each conversation in the dashboard.
• Read up to 20 conversations at once, each with a compact transcript and its labels, to compare them.
• Read one conversation in full, as before.
• Read a conversation's labels: whether it was resolved, why it was flagged for review, its scores, and the messages that justify each answer.
• List the label vocabulary, so the assistant only filters on labels that exist.
• List your sites, with their agents and each agent's live version.

Three ready-made prompts. Your client can offer Analyze a conversation, Compare conversations and Weekly review. Each tells the assistant which tools to use and asks for an answer grounded in message ids, ending with what to change in the agent.

Labels have limits. Labels exist for conversations on the Web, Hosted Page and Playground channels, and a conversation gets them only after it has been quiet for a while. A recent conversation can still read pending, and a label filter can miss it. The server tells the assistant to say so instead of guessing.

Calls are limited per organization. Your organization can make 300 tool calls per minute, shared by everyone connected. Over that, the tool answers with an error asking the assistant to wait 60 seconds. The 300 is a starting value, so rely on the wait time in the message.

What it doesn't give the assistant: shopper identity, reasoning, cost and token usage, or anything outside your organization. Reports and label breakdowns aren't available from MCP yet. It is still read-only: nothing it does changes your agents or your shoppers' experience.

A transcript can contain shopper-provided personal information, so only connect clients you trust with that data. See Connect an MCP client for the tools, the prompts and a worked example.