# Check the schema or data while writing code

> Check the schema or data while writing code: 7 companies fully do it, 5 with a limit, 15 carry the tag. Every listing in Databases & files with the reason and a source.

- Canonical: https://www.anchorterminal.com/niches/data.schema-check/
- Markdown: https://www.anchorterminal.com/niches/data.schema-check/index.md (~1,550 tokens)
- Slim: https://www.anchorterminal.com/niches/data.schema-check/index.min.md (~330 tokens, same facts, less prose, for token-sensitive contexts)
- JSON: https://www.anchorterminal.com/niches/data.schema-check/index.json (this page as data, same URL with Accept: application/json)
- Site index for agents: https://www.anchorterminal.com/llms.txt (the key docs and every listing in one file, at most 16 MB, about 4 million tokens: https://www.anchorterminal.com/llms-full.txt)
- API: https://www.anchorterminal.com/api/v1/index.json
- MCP: https://www.anchorterminal.com/mcp (the whole catalogue as tools, no key)
- DNS: dig +short TXT <slug>.grade.anchorterminal.com (the grade as a DNS TXT record, for a sandbox with no HTTP where names still resolve; also top., best.<category>. and <a>.vs.<b>. under the zone, and the same under grade.letme.dev)
- Licence: CC BY 4.0, credit Anchor Terminal and link this page (https://www.anchorterminal.com/terms/#data)
- Updated: 2026-11-15

Four or more companies' docs show a full fit.

| Companies, full fit | With a limit | Carry the tag | Share of the category's work | First-hand accounts |
| --- | --- | --- | --- | --- |
| 7 | 5 | 15 | 22 | 12 |

## What counts as done

The agent names the tables and columns it needs from the live schema, runs at most a few bounded SELECT or find calls, and writes code that matches them, with no write statement reaching the database.

Measured, for our own tests: the schema lookup for one table takes no more than two calls and under 5,000 tokens.

How agents ask for it: what columns does the orders table have; check the schema before you write the migration; does this user have the right role in the database; verify this query returns what the feature expects

## Does the job

The docs show it meets the test above, and nothing we read says otherwise.

- **Mirage (AgentVFS) (Strukto.AI)**: PostgresVFS mounts read-only with a schemas allowlist; each table has schema.json (types, keys, indexes) and semantic.json, and head or tail on rows.jsonl asks SQL for only those rows, under row and byte caps.
- **ClickHouse MCP Server (ClickHouse, Inc.)**: list_tables takes like filters, page_size and page tokens, and include_detailed_columns=false trims output; read-only is enforced server-side with readonly=1 by default. The 100+ table crash in #25 led to the pagination now documented.
- **DBHub (Bytebase)**: search_objects finds tables by pattern at names, summary or full detail with a limit; read-only is enforced by a keyword allow-list plus the engine's read-only transaction. Two accounts use it for schema checks before writing code.
- **MCP Toolbox for Databases (Google LLC)**: postgres-list-tables takes table_names and a simple or detailed output_format, so one table is one call; read-only sources lock the session at engine level and prune write tools from tools/list.
- **MongoDB MCP Server (MongoDB, Inc.)**: list-collections and collection-schema describe collections; --readOnly leaves only read, connect and metadata tools registered; find and aggregate stop at 100 documents by default; --disabledTools atlas trims the list.
- **Neon MCP Server (Neon, LLC (subsidiary of Databricks, Inc.))**: get_database_tables and describe_table_schema under ?category=schema; ?readonly=true disables writes; run_sql returns 100 rows by default and says when it cut a result; roleName makes grants decide what is read.
- **Supabase API + MCP (Supabase Pte. Ltd.)**: list_tables under ?read_only=true&project_ref=...&functions=database,docs, which drops mutating tools; several accounts use it this way to check schemas and permissions before writing code.

## Does the job, with a limit

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.

- **ObsessionDB (Kellerkind UG (haftungsbeschränkt))**: No schema tool; the agent reads system tables or DESCRIBE through run_sql. The workbench:query scope runs any SQL and no read-only mode is documented, so nothing stops a write reaching the database.
- **PlanetScale MCP Server (PlanetScale, Inc.)**: execute_read_query runs under a read-only database role with short-lived credentials and a 50-second cancel, but returns every row with no cap, and the schema tools sit in the closed, generated part of the server that could not be read.
- **Postgres MCP Pro (Crystal Corp.)**: list_schemas, list_objects and get_object_details find one table's columns, and restricted mode runs read-only transactions, but #237 (open) reports 0.1.1 to 0.3.0 failing to install after an upstream change, and unrestricted is every README example's default.
- **Redis MCP (Redis Ltd.)**: Keys, hashes, JSON and index info can be inspected, but the server has no read-only switch: write and delete tools stay listed, and the README's answer is a read-only Redis ACL user.
- **Snowflake-managed MCP server (Snowflake Inc.)**: A SYSTEM_EXECUTE_SQL tool can read information_schema under the caller's role, but there is no schema tool, and an administrator must first create the MCP server object and an OAuth security integration.

- not this job: CoinMarketCap x402 API (CoinMarketCap): A paid crypto market-data API (quotes, DEX search) with an x402 MCP endpoint; it exposes no schema of the person's database.
- not this job: Filesystem (MCP reference server) (Model Context Protocol, a Series of LF Projects, LLC): File tools only; no database connection.
- not this job: Memory (MCP reference server) (Model Context Protocol, a Series of LF Projects, LLC): A knowledge-graph file store; no database schema or SQL.
- not this job: Nansen x402 API (Nansen): A paid on-chain analytics API (wallets, Smart Money, token flows); it exposes no schema of the person's database.
- not this job: ReqRes MCP server (ReqRes): Works on its own hosted collections, not the repository's database.

## What stops agents finishing

- tool list too large for the client (Cursor sends 40 tools)
- server fails to start on a fresh install after an upstream dependency change
- schema listing crashes on 100+ tables

## Around it

Before: code.understand-codebase. After: code.ticket-to-pr.

## Where this comes from

Weight 22 of 100 in Databases & files: 12 of 52 accounts; the most common job in the DBHub use-case thread and the Supabase MCP comments on Hacker News

As data: https://www.anchorterminal.com/api/v1/use-cases/data.schema-check.json, or the MCP tool get_use_case with id data.schema-check. Draft, read 2026-11-15 from public docs and first-hand accounts, not yet reviewed by a person.

---

Query this catalogue directly over MCP: https://www.anchorterminal.com/mcp (Streamable HTTP, no key). How: https://www.anchorterminal.com/agents/#mcp
