Mercatorgraph

Mercatorgraph turns any codebase into one centralized, queryable knowledge graph — served to many AI agents (via a single MCP server) and to developers (via a graph viewer). It wraps Graphify as the extraction engine: Mercatorgraph consumes Graphify's output, it never forks it.

Why

• 10 developers = 10 duplicated, out-of-sync local graphs. Mercatorgraph builds one central graph per project that every agent and developer reads.
• Graphs die with the session. Here they persist on a volume, versioned.
• No place for human/agent knowledge. Annotations and comments live in Postgres and survive every rebuild — never written into the graph file.

What you get

• A worker that clones a repo, builds the graph, validates it, and promotes it atomically.
• An MCP server exposing fast, scoped, token-authed tools to AI agents.
• A graph viewer (optional) for humans to browse the graph in the browser.

Next

• Getting Started — run the stack and index your first project.
• Architecture — how the pieces fit together.
• Services — worker vs mcp vs view, and why they're separate.
• Deployment — images, registries, release tagging.