Get started — RepoQL

Install the binary

rql is a single self-contained binary — no runtime, no containers, no GPU. It lives quietly beside your IDE.

$ curl -fsSL https://downloads.repoql.ai/latest/install-rql.sh | bash

> irm https://downloads.repoql.ai/latest/install-rql.ps1 | iex

RepoQL speaks MCP. rql install detects the agents on your machine and wires them up — or add it by hand:

$ claude mcp add repoql -- rql mcp

# ~/.codex/config.toml [mcp_servers.repoql] command = "rql" args = ["mcp"]

// .cursor/mcp.json { "mcpServers": { "repoql": { "command": "rql", "args": ["mcp"] } } }

# any MCP client — stdio transport $ rql mcp # or gRPC / CLI: rql query · read · explore

The first tool call your agent makes auto-launches the host. Or start it yourself and watch the codebase come alive:

$ cd your-repo && rql serve

The index builds

Classify, parse, summarize — thousands of files a minute, entirely on your machine.

The dashboard opens

The terrain shows every file arriving. The moment your agent can usefully query, it says so — in your words.

Results carry trust

Every response ends with a footer stating exactly what the index knew when it answered. Honest, or absent.

The first five minutes

minute one

Feel the shape before opening anything — the whole tree, summarized.

read("file:///** => tree: headlines", 5000)

minute two

Ask where something lives. Explore teaches you the repo's real vocabulary.

explore(keywords="auth middleware", question="how do requests get authenticated?")

minute three

Reach for exactly one method body across the codebase — no file opened.

read("file:///src/**#symbol=*Handler.Handle => content", 2000)

minute four

Query the graph like a database — because it is one.

query("SELECT name, uri FROM Types WHERE extends = 'BaseService'")

minute five

Pull a whole other repository into the same index and keep asking.

import("github://dotnet/roslyn")