Novaya-AI/novgraph

Open-source client for Novgraph, Novaya's proprietary hosted code knowledge graph. Gives coding agents recorded intent, related-code context, ranked change impact and architecture context over MCP or

About novgraph

A live knowledge graph of your codebase, queried by coding agents.

Command-line client for Novgraph, a hosted knowledge graph of a git repository. The graph holds your files and symbols as nodes and their real relationships as typed edges, and coding agents query it over MCP or the shell for four things a file read cannot give them: the recorded intent behind each commit, files that change together without importing each other, ranked blast radius, and computed architecture.

Token savings: 78% to 99.74% per answer, measured against the cost of reading the files each answer cites — not modelled, not estimated from a benchmark.

This repository is the client only — auth, agent detection, MCP registration and request transport. Python 3.11+, standard library, no dependencies, no engine. Indexing, storage and retrieval run on Novaya's servers.

install detects every supported agent on the machine, registers the MCP server, writes a pointer into each agent's instruction file, adds a /novgraph command, resolves which indexed codebase this checkout is, and verifies each step. novgraph doctor re-runs those checks with an exit code.

A coding agent starts each session with no memory of the repository. To answer "what does this touch", it greps, opens files, and infers — spending tokens to rebuild a picture it loses at the end of the session. Three classes of fact are not recoverable that way at all:

From the project README.

Maintaining this server?

Add the radar badge to your README — it shows your project was picked up by MCP Radar and links to this page:

[![On MCP Radar](https://mcp.liqiwa.com/badge.svg)](https://mcp.liqiwa.com/s/Novaya-AI--novgraph.html)

Related MCP servers