withcoral/coral

Coral is a local SQL query engine that lets you query multiple APIs and data files as if they were database tables. SQL (Structured Query Language) is the standard way to ask questions of databases, Coral applies that same familiar syntax to cloud services like GitHub, Linear, Datadog, Sentry, and Stripe, as well as local files in formats like JSONL and Parquet. The problem it solves: AI agents typically query each data source separately with its own unique integration, leading to repetitive code, inconsistent handling, and difficulty combining data from different sources. Coral provides one query layer. Each connected service becomes a SQL schema, for example, GitHub issues become the table github.issues, and you can join data across sources in a single query. Coral handles authentication, pagination, and API calls behind the scenes, returning a clean table of results. You can use it directly from the command line (coral sql "SELECT ...") or expose it to AI coding assistants like Claude Code through MCP (a standard protocol for connecting tools to AI agents). The README reports benchmarks showing Claude was 20% more accurate and 2x more cost-efficient using Coral instead of connecting to each provider separately. Custom sources can be defined with YAML configuration files to connect any HTTP API not already bundled. Credentials are stored and used locally only and never sent anywhere beyond the API being queried. Coral is built in Rust and installs on macOS and Linux. The full README is longer than what was provided.