dosco/graphjin

What it solves

GraphJin provides a governed, unified interface for AI agents to interact with a company's existing data systems. It prevents agents from needing raw database credentials or guessing schema structures by exposing databases, warehouses, files, and source code as a single, auditable GraphQL and Model Context Protocol (MCP) surface.

How it works

GraphJin acts as a compiler and runtime that connects to various data sources—including PostgreSQL, MySQL, MongoDB, Snowflake, S3, and local files—and automatically discovers their schemas and relationships. It converts GraphQL queries into optimized SQL or native queries, eliminating the need for manual resolvers or ORMs.

For AI agents, it provides a discovery loop where agents can query a catalog of available data and code before acting. It also includes "CodeSQL," which indexes source code using tree-sitter to allow agents to query the codebase as if it were a database. All actions are checked against an execution ledger and governed by security policies.

Who it’s for

It is designed for developers building AI agents that need secure, read/write access to operational databases, data warehouses, and source code, as well as developers who want a high-performance GraphQL-to-database gateway.

Highlights

• Unified Data Surface: Accesses databases, object stores (S3/GCS), and source code through one GraphQL/MCP interface.
• Built-in Agentic Mode: Includes a server-side agent that can run discovery loops and return evidence-backed answers via a single API call.
• CodeSQL: Allows querying of source code symbols, references, and imports as database tables.
• Governed Access: Features query allow-lists, read-only boundaries, and policy-aware MCP tools to ensure agent safety.
• High-Performance: Compiles requests into single optimized queries to avoid N+1 problems.
• Broad Compatibility: Supports a wide range of backends from PostgreSQL and MongoDB to BigQuery and Snowflake.