qcp-cli 0.1.7

Query Companion: a CLI to query Postgres databases in natural language.

qcp — Query Companion

Query your Postgres database in plain English, right from the terminal.

$ qcp query "what were the top 5 customers by revenue last month?"
Reading schema...
Generating SQL...

SELECT customer_name, SUM(amount) AS revenue
FROM orders
WHERE order_date >= date_trunc('month', now()) - interval '1 month'
  AND order_date < date_trunc('month', now())
GROUP BY customer_name
ORDER BY revenue DESC
LIMIT 5

customer_name   revenue
--------------  -------
Acme Corp       48210.50
Globex Inc      39120.00
...

(5 rows)

qcp uses a LangChain database agent powered by Google Gemini 2.5 Flash to translate your question into a read-only SQL query, run it, print the exact results, and explain the answer in plain language.

Features

• 🗣️ Ask questions in natural language, get SQL + results
• 🔒 Read-only by default — only SELECT/WITH statements are executed
• 💡 qcp insights — AI-generated analytics suggestions for your schema
• 🧠 Local schema memory with automatic 24-hour refresh
• 🔑 Bring your own Gemini API key (free tier available)
• 🪶 Minimal footprint — single CLI, no servers or daemons

Install

QCP requires Python 3.14 or newer.

macOS / Linux (curl)

curl -fsSL https://raw.githubusercontent.com/Moduna-AI/qcp-cli/main/scripts/install.sh | bash

macOS (Homebrew)

brew tap Moduna-AI/qcp-cli
brew install qcp

Windows (PowerShell)

irm https://raw.githubusercontent.com/Moduna-AI/qcp-cli/main/scripts/install.ps1 | iex

Via pip (any platform)

pip install qcp

Quickstart

# 1. Connect a Postgres database
qcp init

# 2. Add your Gemini API key (https://aistudio.google.com/apikey)
qcp auth

# 3. Ask away
qcp query "how many new signups did we get this week?"

# Or get general insights about your data
qcp insights

Running qcp query or qcp insights without a configured database raises a clear error and tells you to run qcp init first.

Configuration

qcp stores config at ~/.qcp/config.json (override with QCP_HOME). Validated schema metadata is cached separately at ~/.qcp/schema.json; it never contains credentials or query result rows. You can also use environment variables instead of (or to override) the config file:

Variable	Purpose
QCP_DATABASE_URL	Postgres connection string
GEMINI_API_KEY	Gemini API key

Check current configuration any time with:

qcp status

Commands

Command	Description
qcp init	Connect a Postgres database
qcp auth	Add/remove your Gemini API key
qcp query "<question>"	Ask a question, get SQL + results
qcp insights	Get AI-generated analytics suggestions
qcp status	Show current configuration

Useful flags on qcp query:

• --dry-run — print the generated SQL without executing it
• --no-show-sql — only print results, hide the SQL

Safety

qcp only executes one SELECT/WITH statement. Validation is backed by a PostgreSQL read-only transaction, and there is no write override. Always review generated SQL (shown by default), especially on production databases. A read-only database role remains recommended as an additional boundary.

Agent tools

The LangChain agent has four narrow tools: lookup_schema, schema_memory, execute_read_query, and analyze_insights. Schema data is reused for 24 hours, then refreshed automatically. An undefined table, column, or schema invalidates the cache and permits one refresh-and-retry cycle.

Development

git clone https://github.com/Moduna-AI/qcp-cli
cd qcp
uv sync --extra dev
uv run pytest

Install the Git hook once with uv run pre-commit install. Use uv run <command> for project commands; uvx creates an isolated environment without QCP installed.

Clean install / clean build

A Makefile (plus equivalent standalone shell scripts) is included so you never get bitten by stale venvs or cached build artifacts:

Task	Make	Script
Wipe venv/dist/caches	make clean	./scripts/clean.sh
Fresh editable install	make install	—
Fresh install + test deps	make dev	—
Clean build (sdist + wheel)	make build	./scripts/build.sh
Clean install, then build	make rebuild	—
Run tests	make test	—
Run the CLI from venv	make run ARGS="status"	—
Run Ruff checks	make lint	—

make build / ./scripts/build.sh always wipe the tree first, so dist/ is never tainted by a previous build. Both are safe to re-run anytime.

Roadmap (post-MVP)

• Additional LLM providers (OpenAI, Claude, local models)
• MySQL / SQLite support
• Query history and caching
• Saved queries / scheduled reports

License

MIT — see LICENSE.

Contributions welcome! Please open an issue or PR.