Reports, label counts and agent versions from your MCP client

Your own Claude, or any MCP client connected to the dashboard MCP server, can now answer questions about how your assistant is doing, not only find conversations. The server goes from six tools to nine.

Read a report. Ask for the conversations, quality, engagement or starter questions report of one site over a range of days, and add the previous period of the same length to compare. The figures come from the same reports as the dashboard, with its default channels (Web and Hosted Page, no Playground). A rate with nothing to divide comes back empty, not as 0%.

Count conversations per label. Ask for the top transfer reasons, objections or resolution outcomes over a period, and add the previous period to see the change per value. The count covers the newest 10,000 labeled conversations of the filter and tells you when it stops there. It also says how many matching conversations have no labels yet, so a count is never mistaken for the whole picture.

Read an agent version. The assistant can read the instructions, the enabled tools and the engine of a version, or what changed between two versions of the same agent. This one depends on your role: it needs Agents: View or Agents: Edit. The permission is checked on every call, so a role change applies at once. A refusal doesn't say whether the version exists.

More ways to search conversations. You can now filter by the type of page the shopper started on (home, category, product or checkout), by product id, by category name, by the engagement widget that opened the conversation, and by commerce outcome. These filters are only in the MCP server, not in the REST API.

Commerce outcomes are unknown for now. Each conversation can say whether the shopper added to cart, reached checkout or purchased, but nothing records those events yet. Until it does, every outcome reads unknown (never "no"), and a commerce filter matches nothing. Once events are recorded for your organization, conversations created from then on get a yes or a no, and earlier ones stay unknown. No revenue or order value is returned.

What it can't answer. The server only knows the conversations your assistant had. It has no site traffic, no revenue and no market or competitor data, and the assistant is told not to estimate any of them. It is still read-only.

The 300 calls per minute limit per organization now covers all nine tools. See Connect an MCP client for the tools, the filters and two worked examples.