Answer a business question from a database or warehouse
One company's docs show a full fit.
What counts as done
A number or small table that answers the question, with the SQL that produced it shown, built from probe queries that each carry a LIMIT or aggregate, and no single result larger than the client can take.
Measured, for our own tests: no tool result above 1 MB and the answer arrives in ten calls or fewer. No docs we read state these figures, so they don't count against a listing here.
How agents ask for it: how much revenue did product X make last month
, which subscriptions look unusual
, top 10 customers by usage this quarter
, chart signups by week
Does the job 1
The docs show it meets the test above, and nothing we read says otherwise.
| Listing | Why |
|---|---|
| Neon MCP Server BB Neon, LLC (subsidiary of Databricks, Inc.) | run_sql returns up to 100 rows per statement by default (max_rows up to 10,000), reports the total when cut off, and returns SQLSTATE detail on errors so the agent can retry. source |
Does the job, with a limit 10
It does the job, but something in the docs or a user's account will get in an agent's way. The limit is named in each row.
| Listing | Why |
|---|---|
| Mirage (AgentVFS) B Strukto.AI | database.json and semantic.json mark dimensions and facts, but reads are file commands with no GROUP BY or aggregate, and the default byte cap is 10 MiB, above the 1 MB bar. source |
| ClickHouse MCP Server BB ClickHouse, Inc. | run_query runs analytical SQL with bound params and EXPLAIN ESTIMATE, but nothing caps rows or bytes; #223 reports a 231 MiB result from one SELECT *, far above the 1 MB bar. source |
| DBHub BB Bytebase | max_rows caps SELECTs and sets truncated: true, and sources carry descriptions, but the cap is rows not bytes, no warehouse engines are supported, and an account reports the model picking the wrong database. source |
| MCP Toolbox for Databases A Google LLC | BigQuery sources cap rows (maxQueryResultRows, default 50) and bytes billed, but postgres-execute-sql has no row cap, no byte cap on results is documented, and #3637 asks for one after 1 MB frames closed a socket. source |
| MongoDB MCP Server A MongoDB, Inc. | find and aggregate cap at 100 documents and 16 MiB by default, above the 1 MB bar unless lowered. #549 shows a date filter wrongly returning 0 documents and #578 the untrusted-data wrapper confusing the model. source |
| ObsessionDB F Kellerkind UG (haftungsbeschränkt) | run_sql caps every result at 1,000 rows or 10 MiB and sets truncated: true, but 10 MiB is above the 1 MB bar, and first connect needs a person to approve OAuth in a browser. source |
| PlanetScale MCP Server BB PlanetScale, Inc. | Read queries go to a replica by default under a read-only role, but the executor returns all rows with row_count and no row or byte cap, so the 1 MB bar rests on the agent's LIMIT. source |
| Postgres MCP Pro F Crystal Corp. | execute_sql runs read-only SQL in restricted mode, but nothing caps rows or bytes per result, so the 1 MB bar depends on the agent's LIMIT. source |
| Snowflake-managed MCP server C Snowflake Inc. | Cortex Analyst returns generated SQL and a Cortex Agent tool answers governed questions, with query_timeout on SQL tools, but access needs an admin-created server object and security integration before any tool works. source |
| Supabase API + MCP BB Supabase Pte. Ltd. | execute_sql answers questions over the project's Postgres, as an account does for bank transactions, but no row or byte cap on results is documented, so the 1 MB bar rests on the agent's LIMIT. source |
Doesn't do this job 6
Listings in the category that do something else, with the reason.
| Listing | Why |
|---|---|
| CoinMarketCap x402 API BB CoinMarketCap | A paid crypto market-data API (quotes, DEX search) with an x402 MCP endpoint; it answers questions about crypto markets, not the person's own warehouse or database. source |
| Filesystem (MCP reference server) B Model Context Protocol, a Series of LF Projects, LLC | No query engine over a database or warehouse. |
| Memory (MCP reference server) B Model Context Protocol, a Series of LF Projects, LLC | No query over a database or warehouse. |
| Nansen x402 API B Nansen | A paid on-chain analytics API (wallets, Smart Money, token flows); it answers questions about crypto markets, not the person's own warehouse or database. source |
| Redis MCP C Redis Ltd. | Key, stream and vector tools only; no SQL or aggregate query over an analytical store. source |
| ReqRes MCP server D ReqRes | Paginated record listing only; no query or aggregate over an analytical store. source |
What stops agents finishing
- unbounded result breaks the transport or the context window
- warehouse needs an admin-created integration or a payment card before tools work
- schema with thousands of tables and no semantic layer
Around it
Before: company-knowledge.find-definition. After: spreadsheets.append-rows, product-analytics.chart.
Where this comes from
Weight 20 of 100 in Databases & files: 11 of 52 accounts (ClickHouse, Supabase finance, MongoDB filters, text-to-SQL and warehouse threads); most vendor pages lead with it
The same as data: /api/v1/use-cases/data.answer-question.json, or the MCP tool get_use_case with id data.answer-question. Other jobs in this category: /api/v1/niches?category=data.
For companies
Do agents find, use and choose your tools?
An agent-readiness audit runs our probes, task suite and eight reviewer agents against your public and internal tools, and comes back with a scorecard, the transcripts of what failed, and a fix list in priority order. From $2,500, re-run included. We never take payment to move a rank. We do help companies earn one.