rocketindex

RocketIndex 🚀

The Deterministic Navigation Layer for AI Agents

One call. Exact answer.

---

Why RocketIndex?

AI coding agents waste tokens because they navigate code by guessing. Grep returns 17 candidates when you need one. LSPs require file coordinates when agents only have a symbol name. Both force agents into multi-turn loops.

RocketIndex replaces guesswork with structure. It parses your code into an AST using Tree-sitter, stores the symbol graph in SQLite, and answers navigation queries deterministically.

Capability	RocketIndex	LSP	Vector Search	Grep
Query model	Symbol name	File coordinates	Natural language	Pattern
Find callers	Yes	Refs only	No	No
Dependency graph	Yes	No	No	No
Precision	Deterministic	Deterministic	Probabilistic	Noisy
Best for	Navigation, refactoring	Type-aware edits	Exploration	Simple edits

vs Grep: Structure beats text matching

Example: Finding spawn in Tokio (751 Rust files)

$ rkt def "spawn"
# -> tokio/src/runtime/blocking/pool.rs:57

Metric	RocketIndex	Grep
Tool calls	1	3-5+
Tokens	~200	5,000+
Result	Exact definition	17 candidates to verify

Grep matches text. RocketIndex understands code structure—it knows the difference between a definition, a call site, and a comment.

vs LSPs: Query by name, not coordinates

LSPs are designed for editors where a cursor is already positioned on a symbol. AI agents don't have cursors—they have symbol names from user requests.

Task	RocketIndex	LSP
"Find UserService.save"	find_definition("UserService.save")	Find file containing text → position cursor → goToDefinition
"Who calls this?"	find_callers("save")	Only "find references" (callers + callees mixed)
"Trace the call graph"	analyze_dependencies("main")	Not available

LSPs also require language runtimes and often fail on incomplete code. RocketIndex uses pure syntactic analysis—it works on partial checkouts, broken builds, and across 15 languages with a single binary.

Built for Scale

Traditional Language Servers choke on large codebases. RocketIndex doesn't.

• Zero Compilation: Pure syntactic analysis. Doesn't care if your code compiles, only that it parses.
• Monorepo Ready: Handles repositories with 10,000+ files.
• Instant Access: SQLite with WAL mode and memory-mapped I/O. Symbol lookups in milliseconds.
• Low Footprint: Targeted disk-based queries instead of loading entire project into memory.

---

Quick Start

Choose your path:

Use Case	Setup
Claude Code	Plugin install (no binary needed)
Claude Desktop, Gemini, Zed	Install binary + configure MCP
CLI for humans/scripts	Install binary + run commands

Claude Code

No install required. The plugin auto-downloads the binary.

/plugin marketplace add rocket-tycoon/rocket-index
/plugin install rocketindex

Install

Required for Claude Desktop, Gemini, Zed, and CLI usage.

macOS

brew install rocket-tycoon/tap/rocket-index

Windows

scoop bucket add rocket-tycoon https://github.com/rocket-tycoon/scoop-bucket
scoop install rocketindex

Linux

curl -LO https://github.com/rocket-tycoon/rocket-index/releases/latest/download/rocketindex-x86_64-unknown-linux-gnu.tar.gz
tar -xzf rocketindex-x86_64-unknown-linux-gnu.tar.gz
sudo mv rkt rocketindex-lsp /usr/local/bin/

Other MCP Clients

After installing, configure your client:

Claude Desktop — add to config file:

Platform	Config Path
macOS	~/Library/Application Support/Claude/claude_desktop_config.json
Windows	%APPDATA%\Claude\claude_desktop_config.json
Linux	~/.config/Claude/claude_desktop_config.json

{
  "mcpServers": {
    "rocket-index": {
      "command": "rkt",
      "args": ["serve"]
    }
  }
}

Gemini CLI:

gemini mcp add rocket-index rkt serve

Zed Editor — add to ~/.config/zed/settings.json:

{
  "context_servers": {
    "rocket-index": {
      "command": "rkt",
      "args": ["serve"]
    }
  }
}

CLI Usage

After installing:

cd /path/to/your/repo
rkt index                    # Build index
rkt watch                    # Keep index fresh (run in background terminal)

Run rkt watch in a background terminal during coding sessions to keep the index fresh.

---

What It Does