# Pawrly

> Query APIs, files, MCP servers, and databases with SQL.

Pawrly lets a team describe each source once in pawrly.yaml, then query APIs, files, MCP servers, and databases with stable table and column names. It works from the CLI, can run as a local service, and can expose the same workspace to MCP clients such as Claude Desktop, Cursor, and Codex.

## When to use Pawrly
Use Pawrly when an agent or script needs data from more than one place — REST/GraphQL APIs, files (Parquet/CSV/JSON), object storage, MCP servers, databases, or warehouses — and you want one SQL question instead of a custom integration per source. It is the read and context path: reach for it before deciding or acting. For a single write to one system (create a ticket, send a message), call that system's own tool directly.

## Agent skills
Pawrly ships a Claude Code + Codex plugin with skills for reading data, connecting sources, saving useful results, and defining approved metrics. See [skill.md](https://pawrly.dev/skill.md).

## Install
- [Install guide for LLMs (Markdown)](https://pawrly.dev/install.md): the fastest path from nothing to a working binary and a first query, copy-pasteable.
- [install.sh](https://pawrly.dev/install.sh): macOS/Linux installer — `curl -fsSL https://pawrly.dev/install.sh | sh`.
- [install.ps1](https://pawrly.dev/install.ps1): Windows PowerShell installer — `irm https://pawrly.dev/install.ps1 | iex`.
- [pawrly.schema.json](https://pawrly.dev/pawrly.schema.json): JSON Schema for pawrly.yaml — reference it with `# yaml-language-server: $schema=https://pawrly.dev/pawrly.schema.json` for editor completion + validation.

## Documentation
- [Quickstart](https://github.com/CITGuru/pawrly#quickstart): install, first query over files, first query over live APIs.
- [Overview](https://github.com/CITGuru/pawrly/blob/main/docs/overview.md): what Pawrly is and how the pieces fit.
- [Sources reference](https://github.com/CITGuru/pawrly/blob/main/docs/sources.md): configure APIs, files, MCP servers, databases, warehouses, and lakehouses.
- [Configuration](https://github.com/CITGuru/pawrly/blob/main/docs/config.md): the pawrly.yaml schema, secrets, and caching.
- [MCP](https://github.com/CITGuru/pawrly/blob/main/docs/mcp.md): run Pawrly as an MCP server and consume other MCP servers as sources.
- [Semantic layer](https://github.com/CITGuru/pawrly/blob/main/docs/semantic.md): approved fields, metrics, joins, access rules, and required filters.
- [Materialize](https://github.com/CITGuru/pawrly/blob/main/docs/materialize.md): save a query result, file, or URL as a reusable table.
- [CLI](https://github.com/CITGuru/pawrly/blob/main/docs/cli.md): sql, schema, validate, serve, status, mcp-stdio.
- [Architecture](https://github.com/CITGuru/pawrly/blob/main/docs/architecture.md): implementation details and local vs daemon behavior.
- [Observability](https://github.com/CITGuru/pawrly/blob/main/docs/observability.md): metrics and query auditing.

## Writing
- [Agents Need a Query Surface, Not More Tools](https://pawrly.dev/blog/agents-need-a-query-surface-not-more-tools): I keep coming back to the same problem with agents.

## Optional
- [GitHub repository](https://github.com/CITGuru/pawrly): source, issues, and releases.
- [Example configurations](https://github.com/CITGuru/pawrly/tree/main/examples): a kitchen-sink workspace covering every source kind.