What it does GraphJin is a single binary that connects to your database, reads the schema, and exposes it as an MCP (Model Context Protocol) server. Claude, GPT-4, or local 7B models can then ask natural-language questions that get compiled to optimized SQL. It supports PostgreSQL, MySQL, MongoDB, SQLite, Oracle, MSSQL, Snowflake, BigQuery, and even S3/files.

The interesting bit The “CodeSQL” feature indexes your source tree with tree-sitter into a SQLite cache, then links code symbols to database columns. An agent can ask “which handlers touch customer invoices?” and get actual file paths, not hallucinated guesses. That’s the rare case where gluing two boring things together—AST parsing and schema introspection—actually becomes useful.

Key highlights

• Auto-discovers foreign keys, relationships, and indexes; no resolvers or ORM needed
• Compiles GraphQL-ish queries to a single optimized SQL statement, avoiding N+1
• Built-in OIDC device-code auth with JWT audit logging
• Guided MCP installer for Claude Code and OpenAI Codex
• Analytics directives: running totals, moving averages, window functions via GraphQL syntax
• Real-time subscriptions (details truncated in README)

Caveats

• The README is enthusiastic about “no configuration required,” but production use clearly needs YAML for sources, auth secrets, and CodeSQL paths
• MongoDB and older databases get “clear errors” for analytics directives—supported SQL databases validate at compile time, others don’t
• The “thousands” claim for real-time subscriptions is cut off; actual throughput numbers aren’t visible

Verdict Worth a look if you want Claude to actually query your production database without hand-writing SQL or maintaining a separate API layer. Skip it if you already have a mature GraphQL server with custom resolvers you trust.