Travsr
The code graph that lives next to git.
Travsr is a graph-native, always-fresh code intelligence daemon. Instead of treating your codebase as unstructured text, Travsr builds a live graph and exposes it via MCP so AI agents traverse the graph rather than guessing from chunks.
Result: 80% fewer tokens, zero structural hallucinations, always-fresh context.
Quick Start
npm install -g travsr
cd your-repo
travsr init
How It Works
- Indexing. Tree-sitter parses your code into a structural graph. LSIF compilers (tsc, rust-analyzer, pyright) add semantic edges including call graphs, type references, and data flow.
- Storage. The graph lives in SQLite (MVP) or Kùzu (production). Node identity uses Kythe VNames for cross-repo stability.
- Retrieval. BFS, Personalized PageRank, and PCST algorithms return the minimal subgraph relevant to a query.
- Serving. The graph is exposed via the Model Context Protocol over stdio (local) or SSE (cloud).
Supported Languages
| Language | Tree-sitter | LSIF (semantic) |
|---|---|---|
| TypeScript | S1 | S4 |
| Rust | S8 | S9 |
| Python | S10 | S13 |
| Go | S12 | S13 |
MCP Tools
| Tool | Description |
|---|---|
get_dependencies | Direct and transitive imports for a file |
get_callers | All call sites of a symbol across repos |
get_blast_radius | What breaks if this file changes |
get_execution_path | PCST path between two symbols |
get_context | Full PPR + knapsack context for a query |
search_symbol | Find symbol definitions by name |
get_repo_map | Structural overview of a repository |
Architecture
Travsr is a Rust workspace with 7 crates:
travsr-core Graph engine: nodes, edges, Kythe VNames
travsr-indexer Tree-sitter + LSIF parsing pipeline
travsr-store Storage abstraction (SQLite / Kùzu)
travsr-retrieval BFS, PPR, PCST, k-core, knapsack
travsr-mcp MCP server (stdio + SSE)
travsr-daemon Git hook, file watcher, scheduler
travsr-cli travsr binary entrypoint