cb-analytics-querylisted
Install: claude install-skill celticht32/Couchbase-Skills-for-Claude.ai
# Querying Couchbase Analytics via cb-analytics-mcp
You have **five SQL++ tools** for talking to the Analytics service. Picking
the right one matters — they have different cost profiles, rate-limit
budgets, and result shapes.
| Tool | Cost | Result shape | When to reach for it |
|---|---|---|---|
| `execute_query` | high (full result) | rows, may be truncated | DDL, mutations, small SELECTs |
| `execute_query_readonly` | high (full result), **cached** | rows, may be truncated, `cached: bool` | repeated SELECTs in an investigation loop |
| `execute_query_paginated` | constant per page | first page + handle | SELECTs that might return many rows |
| `fetch_next_page` | constant per page | next page for a handle | follow-up to paginated |
| `explain_query` | tiny (no execution) | query plan | "why is this slow" |
## The cardinal rule: pick the tool that matches what you'll do with the result
If you only need the **first N rows** to answer the user's question, use
`execute_query_paginated` with `page_size=N`. Don't pull a million rows
through the MCP boundary just to take the first 20.
If you need to **show the user the data** (and the dataset is small),
`execute_query_readonly` is fine — but watch for `truncated: true` in the
response.
If you need to **make a decision** based on aggregates (`COUNT`, `SUM`,
`AVG`, `GROUP BY`), the result is small by construction. Use
`execute_query_readonly` and benefit from the cache.
## The soft cap (you will see `truncated: true`)
Both