coral

Coral is a local, single query interface designed for AI agents to access data from various sources like GitHub, Slack, Datadog, Linear, and local files. Instead of making multiple tool calls per source, agents write SQL queries that Coral translates into API calls or file reads, returning a single tabular result set. This approach reduces token usage, simplifies cross-source joins, and improves efficiency for complex read tasks. Key features include: - **SQL over APIs**: Query multiple live sources through SQL, with tables exposed as schemas (e.g., `github.issues`). - **Cross-source joins**: Execute joins across different sources in one statement, with local execution after fetching data. - **Local-first**: All data, credentials, and usage history stay on the user's machine. - **MCP integration**: Expose Coral to agents like Claude Code, Codex, Cursor, or VS Code via a built-in MCP server. - **Desktop and CLI**: Available as a desktop app for macOS, Linux, and Windows, plus a CLI for servers and automation. - **Extensible sources**: Bundled sources (Datadog, GitHub, Linear, Sentry, Slack, Stripe, etc.) plus community specs and custom source definitions in YAML. - **Production-ready**: Handles pagination, caching, and query pushdown to minimize API traffic. Benchmarks against direct provider MCPs show improved accuracy and cost efficiency, especially for complex coding agent tasks. The project is Apache 2.0 licensed and actively developed with community support via Discord and GitHub.