internal/

Package bench measures the project's core bet — token efficiency — the same headline the upstream codebase-memory-mcp reports (10x fewer tokens, 2.1x fewer tool calls on structural questions; see docs/UPSTREAM.md).

Package gocalls resolves Go CALLS edges in-process via go/packages + a VTA (Variable Type Analysis) call graph — the same machinery gopls uses, no subprocess, no main needed.

Package graph defines the knowledge-graph data model and its SQLite store.

Package install registers the codegraph MCP server into the AI coding agents found on the machine.

Package mcp implements a minimal Model Context Protocol server over stdio.

Package quality is the answer-quality harness — the half of the upstream benchmark our token harness deliberately left out.

Package query turns store calls into compact, agent-friendly results.

Package scip maps a SCIP index (produced by scip-typescript) into codegraph CALLS edges.

Package similar finds near-clone code via MinHash over token shingles — the M4 SIMILAR_TO pass.