MCP Tool
get_project_index
The first tool the AI should reach for. Always.
Purpose
Returns a compact map of the entire project, file paths, symbols, imports, at a chosen detail level.
Parameters
| Name | Type | Required | Description |
|---|---|---|---|
level | 1 | 2 | 3 | 4 | No (default 2) | Detail level, see below |
folder | string | No | Restrict to files under a path, e.g. src/auth |
Levels at a glance
| Level | Includes | Tokens / file | Use when |
|---|---|---|---|
| 1 | Paths + language | ~8 | "What files exist?" |
| 2 | + symbols + imports | ~80 | General exploration (default) |
| 3 | + full signatures | ~150 | Before refactoring |
| 4 | + AI summaries | ~250 | Architectural understanding |
Example calls
lightweight overview
{ "level": 1 }focused exploration
{ "level": 2, "folder": "src/auth" }pre-refactor analysis
{ "level": 3, "folder": "src/db" }Sample output (level 2)
response
# larkx context · level 2 · ~340 tokens · 4 files
src/auth/login.ts[ts]: validateJWT@6, hashPassword@10, LoginService@14 | +../utils/crypto
src/auth/middleware.ts[ts]: authMiddleware@11, rateLimiter@20 | +./login
src/auth/session.ts[ts]: createSession@4, destroySession@14 | +./login
src/auth/index.ts[ts]: | +./login,+./middleware,+./sessionAlways start hereCalling
get_project_index first is almost always cheaper than reading files directly. If the answer isn't visible in level 1 or 2 output, call again at a higher level or use a more specific tool.See: Levels deep dive