Search symbol DECLARATIONS by name/substring across the project via the language server's index (clangd C/C++, Roslyn C#/.NET, tsserver JS/TS, pyright Python — auto-detected) — NOT grep. Returns a token-capped kind name @ file:line list, no source bodies. Use this instead of Bash grep/rg for finding a class/function/type/variable in any of those languages.

Find every call site / usage of a symbol (semantic, via the language server) — NOT a text grep. THE tool to reach for when MODIFYING code: pass symbol (just the name, e.g. "SpawnActor") and it resolves the declaration and returns all references in one call — no need to know a line/column. (A 0-based path+line+character position also works, to disambiguate an overload.) Returns a token-capped file:line list, no bodies. Prefer this over grepping a name to find its uses.

Resolve the definition of the symbol at a 0-based position (semantic, via the language server). Returns a token-capped file:line list, no bodies.

Type/signature info for the symbol at a 0-based position (language-server hover). A few plaintext lines, no walls of docs. Use to check a type/overload without opening the file.

Outline a single file: its classes/functions/types as a token-capped kind name @ file:line list. Cheaper than reading the whole file to see its structure.

Semantically rename the symbol at a 0-based position across the whole project (language-server rename — updates every reference, never a text sed). Default is a PREVIEW returning the affected file:line list; pass apply=true to write the edits to disk. Use this instead of editing call sites by hand.

Replace a whole declaration (function/method/class body, signature included) by NAMING it — the language-server outline supplies the exact span, so you don't Read the file into context or count lines for an exact-match Edit. Default PREVIEW returns the affected file:line; apply=true writes. Token-cheap symbol-level editing instead of read-whole-file-then-Edit.

Insert text on a new line AFTER a named declaration (e.g. add a sibling function/method). The outline supplies the insertion point. Default PREVIEW; apply=true writes. No need to Read the file to find the line.

Insert text on a line BEFORE a named declaration (e.g. add an import/attribute/decorator above it). The outline supplies the insertion point. Default PREVIEW; apply=true writes.

Delete a named declaration — but REFUSE while it's still referenced (so a delete can't silently orphan call sites). Checks references first; lists them and stops unless force=true. Default PREVIEW; apply=true writes.

Find files by name (substring or glob like *Manager.cpp) under the project root — token-capped file list. The sanctioned replacement for Bash find -name. No language server needed.

Raw text/regex search in source (string literals, comments, config keys — things the symbol index can't answer). Bounded and token-capped to file:line: trimmed-line. The sanctioned replacement for Bash grep when you genuinely need text, not symbols. Prefer search_symbol for code symbols.

Run a READ-ONLY git command and return its output COMPACTED (token-capped) — for status/log/diff, which the language-server index can't help with but whose raw dump is verbose and repetitive. status groups by change-type + directory; log keeps one line per commit; diff collapses to a per-file +/- diffstat (no hunk bodies). Mutating subcommands (commit/reset/checkout/clean/push/merge/rebase) are REFUSED — this only compacts output; run those directly. Use instead of a raw git status/log/diff.

Run a READ-ONLY Perforce (p4) command and return its output COMPACTED (token-capped) — for opened/status/reconcile/changes, whose raw output is long and repetitive. Groups files by action + depot directory and caps the list. reconcile is forced to preview (-n); mutating subcommands (submit/revert/edit/add/delete) are REFUSED — run those directly. Use instead of a raw p4 opened etc.

Configure vs-token-safer (projectPath, backend, maxResults). Writes ~/.vs-token-safer/config.json; run /reload-plugins after. Precedence: env (VTS_*) > config file > default. Can also generate the C++ compile DB in the same step (genCompileDb).

Show current effective vs-token-safer settings + config-file path.

Report how many tokens you've saved vs forwarding raw index responses (local, cumulative). Optional graph/daily/history breakdowns + an estimated USD value.

Clear the local savings ledger.

Scan recent Claude Code transcripts (local, read-only) for code searches that BYPASSED vts — Bash grep/rg/find or the Grep tool aimed at source — and report the raw tokens they spent (the missed savings). Use to see where token-heavy text search is still slipping past vts.

Pre-build the language-server index (IDE-style) so later searches are fast. Spawns + warms the backend without running a query.

Generate compile_commands.json for an Unreal project (so clangd gets a full semantic index) by running UBT GenerateClangDatabase. The user's choice vs staying in no-DB text mode. DRY RUN by default (prints the exact command); apply=true runs it (takes minutes, needs the UE build env). The DB and clangd's .cache/ index land OUTSIDE the source tree (~/.vs-token-safer/db/) so git/p4 never see them; inTree=true keeps the classic project-root layout (then a VCS-ignore guard protects it).