# Ask an AI assistant about your analytics

With the [Lettermint MCP server](/mcp), an AI assistant such as Claude, ChatGPT, or Cursor can query your analytics. You ask a question in plain language. The assistant builds the query, runs it, and explains the result.

The assistant uses the tool `query_email_analytics`. This tool runs the same query as the [API](/platform/analytics/query) and returns the same numbers.

## Before you start

You need:

- A team on the Pro plan.
- An AI client that is connected to the MCP server. See [Connect your client](/mcp#connect-your-client).
- A team that allows MCP access.
- A [role](/platform/teams/roles) with the **View analytics** permission.

The MCP server uses your Lettermint account. You do not need a Team API token.

## Check that analytics are available

Ask the assistant for your teams:

> Use Lettermint to list my teams. Show the capabilities of each team.

The assistant calls `list_teams`. A team that you can query has `query_email_analytics` in its capabilities.

If the capability is missing, one of these conditions applies:

| Condition | Solution |
| --- | --- |
| The team is not on the Pro plan. | Upgrade the team. |
| Your role does not have **View analytics**. | Ask an owner or an admin of the team to change your role. |
| The team does not allow MCP access. | A member with **Manage team security** can turn it on in [**Manage team > Settings**](https://app.lettermint.co/team/settings). |

## Ask a question

A good question names four things. The assistant then has less to guess.

| Part | Example |
| --- | --- |
| The team | "for the team Acme" |
| The metric | "the bounce rate" |
| The window | "of the last 7 days, in Amsterdam time" |
| The split | "for each mailbox provider" |

> For the team Acme, show the bounce rate and the deferral rate of the last 7 days in Amsterdam time, for each mailbox provider. Sort by deferral rate.

You do not need the exact metric names. The assistant reads them from the tool. If you know a name from the [metrics](/platform/analytics/metrics) page, use it. An exact name removes doubt.

### Example prompts

Find a problem:

- "Which mailbox providers deferred our email yesterday? Show the number of deferrals and the deferral rate."
- "What are the ten most frequent SMTP replies behind our bounces of this week?"
- "Show the delivery rate for each hour of today. Tell me when it dropped."

Compare:

- "Compare the bounce rate of the last 7 days with the 7 days before. Show the change in percentage points."
- "Compare the complaint rate for each value of the tag `campaign`, for this month."

Measure speed and engagement:

- "How long does delivery take? Give the median and the 99th percentile for each provider, for the last 24 hours."
- "For our broadcast email of last week, how many recipients had a human open, and how many opens came from privacy proxies?"

## Check the answer

An assistant can read a correct result and still explain it incorrectly. Ask for the evidence:

> Show the query that you sent, the window from `meta`, and the rate bases.

Then check these points.

| Point | Why |
| --- | --- |
| The window | An assistant sometimes selects a different window than you meant. `meta.from` and `meta.to` are the facts. |
| The time zone | The default is UTC. If you want local days, say so. |
| The counts behind a rate | A rate from a small number of recipients tells you little. The counts are in `rate_bases`. |
| Fractions and percentages | The tool returns rates as fractions. `0.02` is 2 percent. |
| "Delivered" | A delivered email was accepted by the receiving server. It is not proof that the email is in the inbox. |
| Sums of breakdown rows | A breakdown by provider counts delivery attempts and leaves out the group without a value. The rows can add up to less than the total. See [Dimensions that change what is counted](/platform/analytics/dimensions#dimensions-that-change-what-is-counted). |
| Sums of open classes | The classes `human`, `privacy`, `machine`, and the others overlap. Their sum is not the total. |

## Two tools return numbers

The MCP server has two tools for statistics. They use different rules, so their numbers are different.

| Tool | Returns | Use it for |
| --- | --- | --- |
| `query_email_analytics` | Deliverability analytics, as this section describes | Rates, breakdowns, filters, latency, and comparisons |
| `get_delivery_stats` | Daily totals for each project, as the dashboard statistics show them | A count that must match the dashboard, and inbound email |

Do not let the assistant add or compare the numbers of the two tools. If you want deliverability analytics, say "use `query_email_analytics`" in your prompt.

When analytics are not available for a team, the assistant must tell you. It must not replace the answer with `get_delivery_stats` without telling you.

## What the assistant can see

- **Your projects only.** The tool reads the projects in your [project access](/platform/projects-and-routes/project-access). Two members of one team can get different totals for the same question.
- **No recipient addresses.** Analytics do not store them.
- **No from address or subject by default.** These two dimensions hold personal data. They are available only when personal-information filtering is off for the team, and your role has **View messages**. See [Personal-information filtering](/mcp#personal-information-filtering).

The tool reads data. It cannot change your team, your projects, or your email.

## Limits

The MCP server and the API share the query limits of a team: 60 queries in one minute, and 2 at the same time. One question can need several queries. A window can be 90 days at most.

See [Limits](/platform/analytics/query#limits) for the full list.

## Troubleshooting

| Problem | What to check |
| --- | --- |
| The assistant says that analytics are not available. | Check the plan of the team and the **View analytics** permission. |
| The assistant uses `get_delivery_stats`. | Ask for `query_email_analytics` by name. Check that the team has this capability. |
| A query on subject or from address fails. | Check personal-information filtering and the **View messages** permission. |
| The tool reports a combination that is not possible. | Not every metric works with every dimension. See [Which dimensions work together](/platform/analytics/dimensions#which-dimensions-work-together). |
| The tool reports that the query reached the time limit. | Ask for a shorter window or fewer splits. |
| The tool reports too many queries. | Wait one minute. Ask one question at a time. |
| The numbers differ from the numbers of a colleague. | Compare your project access. |

## Next steps

- [MCP server](/mcp). Connect a client and review how access works.
- [Metrics](/platform/analytics/metrics). Learn the exact names and definitions.
- [Examples](/platform/analytics/examples). See the query behind each common question.
