SQffy is a simple command-line utility for working with SQL databases from the terminal. Connect, write a query, read the answer.

SQLite, PostgreSQL, MySQL, MariaDB, ClickHouse, and Oracle are built in. Any other database can be added as an adapter: same shell, same commands. The checklist is in docs/DRIVERS.md.

Fast at the keyboard

SQffy is written in Rust, and the shell stays quick. Syntax highlighting paints the query while you type. Autocomplete brings tables and columns onto the line, so you spend the time on the data, not on remembering names.

After FROM, the list is tables. After the table name, it is columns.

Tables that are easy to read

Schema and query results are rendered, not dumped as raw text. The type sits under the column name — the type the engine reported. .schema puts columns, keys, and indexes on one screen, plus the DDL when the engine returns it.

Need rows for a test or a demo? .generate fills a table: data generation for testing and visualization. A preview prints first. --yes inserts. A hundred rows or more without --yes is refused.

Any terminal

Windows, Linux, and macOS. A normal terminal is the whole interface — the one you already have open.

A safe MCP server for AI agents

Modern work often includes an agent in the editor. SQffy speaks MCP, and that server is the safe door.

The agent can inspect the schema and run a read: SELECT, WITH, SHOW, EXPLAIN, PRAGMA. A dangerous statement does not reach the database. INSERT, UPDATE, DELETE, DROP, and the rest of the writes are refused on this server only. At the prompt, the same statements run as you write them. HTTP listens on 127.0.0.1.

sqf mcp                      # stdio, for an editor or agent
sqf mcp --port 8787          # http://127.0.0.1:8787/mcp   ·   GET /health
sqf --mcp-port 8787 ./app.db # the same server, beside the prompt

An easy start

A database comes up in a couple of steps, even when the engine is not installed yet. Run sqf and choose create.

If nothing is listening locally, SQffy offers to pull a Docker image and run it, or to install the package for you (winget on Windows). When neither path fits, it opens the official download page. You do not have to arrive with Postgres already on the machine.

The picker has three moves:

1. Connect to a saved database
2. Add one you already have
3. Create a new one — a file, or a server

sqf                              # picker
sqf ./app.db                     # straight to a SQLite file
sqf postgres://USER:PASSWORD@localhost:5432/app

.tables
.schema users
SELECT id, name FROM users WHERE id >= 5;
.generate users 1000 --dry-run

Connections are remembered in connections.toml. --no-save leaves the file alone.

Tests can point the catalog elsewhere with SQFFY_HOME.

Install

Rust 1.85 or newer. From a checkout of this repository:

cargo install --path .

# or run it without installing
cargo run --release --bin sqf

SQLite ships inside the binary. Oracle needs Instant Client on the machine.

sqf --lang ru                   # en · ru · zh
SQFFY_LANG=zh sqf

Language resolution: --lang, then SQFFY_LANG, then LC_ALL, LC_MESSAGES, LANG, then English. Strings live in docs/I18N.md.

cargo build --release --no-default-features --features sqlite

Reference

Meta commands

Flags

Databases

MCP tools

Docs

cargo test       # unit tests
cargo e2e        # CLI end-to-end: SQLite, MCP, meta commands
cargo e2e-all    # plus server adapters — set SQFFY_E2E_* first

docker compose -f e2e/docker-compose.yml up --build --abort-on-container-exit

sqffy/
├── src/        CLI, drivers, REPL, MCP, generate, provision
├── locales/    en.toml · ru.toml · zh.toml
├── docs/       drivers, i18n, screenshots
├── tests/      e2e_*  (feature = e2e)
├── e2e/        Docker Compose
└── .github/    CI