Skip to main content

Frequently Asked Questions

General​

Is Travsr free?​

Yes. The local tier is Apache-2.0-licensed open source and will always be free. Cloud and Enterprise pricing will be announced when those tiers open; the roadmap tracks where those tiers have got to.

Does Travsr send my code anywhere?​

Not by default. The local daemon indexes your code on your machine and stores the graph in .travsr/graph.db. Nothing leaves your machine unless you explicitly connect a repository to Travsr Cloud.

Which languages are supported?​

Travsr parses 16 languages structurally with Tree-sitter out of the box, with zero configuration: TypeScript/TSX, JavaScript, Rust, Python, Go, Java, Kotlin, Scala, C#, Ruby, PHP, C, C++, Swift, Objective-C, and Dart.

Semantic call edges for TypeScript / JavaScript, Rust, and Python are built into the binary and run automatically. The remaining languages ship as Phase B indexers, installed on demand: Go, Ruby, PHP, C, C++, Swift, Objective-C, Dart, Java, Kotlin, Scala, and C#.

travsr lang detect # scan the repo and auto-install what it finds
travsr lang install java # or install a specific language
travsr lang list # show installed indexers and their status

All indexers extract structural edges via Tree-sitter (import edges, definitions, and call-expression references), which power every MCP tool and retrieval query.

How is Travsr different from GitHub Copilot / Cursor?​

Most AI coding tools fall back on text search: grep/ripgrep for exact matches, vector RAG for fuzzy ones (chunking your code, embedding the chunks, and retrieving the closest match at query time). Both are approximate, and vector RAG is often stale by design.

Travsr builds a deterministic graph from your code and answers structural questions with graph algorithms (BFS, PPR, Dijkstra path search), not pattern matching. The graph is always fresh because the daemon updates it on every save and Git hooks reconcile every commit. For natural-language queries, a small local cross-encoder reranks the graph's own candidates and reports honest confidence instead of guessing; vector embeddings, when installed, are an opt-in recall booster, not the relevance signal. See Why graph beats text search for a detailed comparison.


Setup​

The daemon won't start. What do I check first?​

  1. Is travsr on your PATH? Run which travsr.
  2. Is the socket in use? rm -f /tmp/travsr.sock && travsr daemon start.
  3. Check logs: travsr daemon logs --lines 50.

Do I need to run travsr init on every repository?​

Yes, once per repository. After init, the running daemon reindexes files as you save them and the git hook handles incremental reindexing automatically on every commit.

How do I update Travsr?​

npm update -g @travsr.com/travsr # if installed via npm

Graph & indexing​

How long does the first index take?​

It scales with repository size. The Tree-sitter (Phase A) structural pass is the fast part; semantic Phase B (LSIF/SCIP) is slower and depends on type complexity. Subsequent commits only reindex changed files.

What is the maximum repository size?​

SQLite is the zero-setup default and handles everyday repositories and large monorepos. A production-scale graph database backend for very large graphs is planned (see the Roadmap).

Can I index a monorepo?​

Yes. Run travsr init at the monorepo root. The indexer traverses all subdirectories and respects .gitignore. Cross-package edges within the monorepo are fully resolved.

Does Travsr index node_modules / target/?​

No. node_modules, target/, .git/, and any patterns in .gitignore are excluded automatically.


MCP & Claude​

What MCP tools does Travsr expose?​

Twenty-eight in a single-repo stdio session (travsr mcp), seventeen in a global stdio session (travsr mcp --global), and seven over SSE (travsr serve).

  • Single-repo stdio (28): the graph query and observability tools search_symbol, get_dependencies, get_callers, find_references, find_pattern, get_blast_radius, get_execution_path, get_repo_map, get_context, get_graph_stats, get_graph_json, get_snippets, get_lang_status, get_architecture_brief, get_subsystem_brief, get_index_status, get_graph_health, get_daemon_logs, and repo_languages; the six synonym_* tools (synonym_add, synonym_set, synonym_remove, synonym_remove_term, synonym_list, synonym_reset); and the registry tools repos_list, repos_prune, and repos_remove.
  • Global stdio (17): the same set minus get_architecture_brief, get_subsystem_brief, repo_languages, the six synonym_* tools, repos_prune, and repos_remove. repos_list stays.
  • SSE (7): only get_dependencies, get_callers, get_blast_radius, get_lang_status, search_symbol, get_repo_map, and get_execution_path. The SSE server has no tools/list; clients call these by name.

Graph query results that carry repository content are sanitized and wrapped in a <travsr-data> envelope before they reach a model. A few tools return raw structured output instead: get_graph_stats, get_graph_json, get_lang_status, repo_languages, and the synonym_* and repos_* tools. See MCP Tools Reference for full schemas.

Claude Desktop doesn't see Travsr as a tool. What's wrong?​

  1. Verify your claude_desktop_config.json has valid JSON (a syntax error silently disables all MCP servers).
  2. Check that travsr mcp --stdio launches without errors: run it manually in a terminal.
  3. Fully quit and relaunch Claude Desktop: the MCP config is only read at startup.

Can I use Travsr with Cursor / Continue / other agents?​

Yes. Any MCP-compatible agent works with the stdio transport. The travsr serve command also exposes an SSE transport for clients that support the MCP SSE protocol.