Troubleshooting

A jarvis tool returned empty or an error. Start here.

Start here: the three most common failures

Symptom	Entry
First MCP connect never finishes	First MCP connect times out (uvx cold start)
`jarvis`/`jarvis-server` command not found	jarvis or jarvis-server not found (PATH)
`jarvis index` refuses to run	jarvis index refuses an old scip (version gate)

(documentSymbols, goToDefinition, findReferences, callHierarchy, typeHierarchy)

Run jarvis status <slug>. Is the repo indexed?
- Not indexed → run jarvis index <slug> and try again.
- Indexed → continue.
Does status show SCIP coverage or search-only fallback?
- Search-only → the language indexer produced zero shards. Install the right language indexer via setup.sh and reindex. See Common Failures: navigation empty but search works.
- SCIP coverage → continue.
Is it typeHierarchy specifically?
- Yes → upstream scip v0.9.0 doesn’t populate relationships. Ensure the fork build is installed. See Upstream Issues: typeHierarchy.
- No → check the symbol name. Bare names that collide across packages return candidates. Try a qualified name.

(searchCode, semanticSearch)

Is it semanticSearch?
- → Check the [semantic] extra is installed, and that the second MCP server is running a Python that ships semantic wheels (>=3.12). See Common Failures: semanticSearch error and semanticSearch server fails to connect.
Is it searchCode?
- → Check the Zoekt repo name pin. See Common Failures: wrong repo / missing repo.

→ Reindex: jarvis reindex <slug>. See Common Failures: stale.

→ This is by design (no per-node timestamp on the graph). See Common Failures: blastRadius unknown.

If the troubleshooting steps above don’t resolve your issue, ask on GitHub Discussions — the maintainers and community can help.