Travsr Documentation

The code graph that lives next to git.

Travsr is a graph-native, always-fresh code intelligence daemon. It builds a live deterministic graph of your codebase on every commit and exposes it via Model Context Protocol (MCP) so AI agents traverse real structure instead of guessing from text chunks.

Get started in 5 minutes

New to Travsr? Follow the Quickstart to go from zero to your first MCP query in Claude Desktop.

Guide	What you'll do
Quickstart	Install, index your repo, connect Claude Desktop
Installation	All install methods: npm, binary, source
Architecture	How Tree-sitter → LSIF → MCP fits together

Reference

Section	Contents
CLI Reference	Every travsr subcommand with flags and example output
MCP Tools	All 12 tools with full JSON Schema and examples
VS Code Extension	Features, settings, troubleshooting

How it works

Your repo
  │
  ├─ git post-commit hook
  │      └─▶ travsr hook-run (incremental)
  │
  ├─ Tree-sitter     → structural AST edges (all languages)
  ├─ LSIF compiler   → semantic edges (call graph, type refs, DFG)
  └─ Kythe VNames    → globally stable node identity
          │
          ▼
    SQLite graph
          │
          ▼
    BFS → PPR → PCST retrieval
          │
          ▼
    MCP stdio (local) or SSE (cloud)
          │
    Claude Desktop / Cursor / any MCP agent

Why not vector RAG?

Source code is a deterministic mathematical graph: not unstructured text. Chunking and embedding it destroys the structure that matters. Travsr answers structural questions with graph algorithms, not approximate similarity. See Why graph beats RAG for the full comparison.