Give Claude Code
Structural Superpowers

Real-time incremental code graph with 15 native MCP tools. Your AI assistant finally understands how your code connects.

0
MCP Tools
0.00ms
Per Update
0+
Tests Passing
0
Languages
The Problem

Claude Code is powerful, but flies blind on code structure

Answering "what breaks if I change this?" burns 400+ tokens
Grep is blind to semantic relationships between files
Multi-turn exploration wastes context window and budget
Dependency tracing takes 80+ seconds of guesswork
The Solution

A live code graph that updates in 0.05ms

Incremental updates on every edit — never stale
Pull-based tools — Claude decides when to use them
Native AST parsers with 3-tier fallback chain
Zero config — activates automatically on first edit
Live Simulation

Watch your code graph come alive

StreamRAG traces dependency chains in real-time. See how analyze_impact illuminates blast radius across your entire codebase.

function
class
module
import
calls
imports
inherits
uses_type
Live Terminal Demo

See StreamRAG in action

Real tool outputs. Real data. Zero exploration overhead.

claude code — streamrag
LIVE
15 Native MCP Tools

Every tool Claude needs to understand your code

Pull-based architecture means Claude calls tools on demand. No noise, no overhead — just instant structural answers.

Navigation

5 tools
find_symbolLocate where any symbol is defined
get_callersFind who calls or imports a function
get_calleesDiscover what a function calls
get_file_depsTrace forward file dependencies
get_file_rdepsReverse dependency analysis

Analysis

4 tools
analyze_impactFull blast radius of any change
find_pathShortest dependency chain A to B
find_dead_codeDetect unreachable code
find_cyclesCircular dependency detection

Overview

3 tools
scope_taskExact files needed for a task
get_architectureHigh-level module map
get_file_structureAll entities & relationships

Smart

3 tools
ask_graphNatural language queries (70+ patterns)
session_changesTrack what changed this session
visualize_depsMermaid/DOT dependency diagrams
System Architecture

Built for real-time performance

Five layers working in concert. Native daemons for speed, incremental updates for freshness, and a pull-based architecture for zero overhead.

MCP Server

JSON-RPC stdio transport to Claude Code

JSON-RPC ProtocolDaemon FallbackUnix Socket IPC

15 Native MCP Tools

Pull-based tools Claude calls on demand

Navigation (5)Analysis (4)Overview (3)Smart (3)

Incremental Graph Engine

0.05ms per update via semantic diffing

AST ExtractionDelta ComputationSurgical PatchingEdge Resolution

Three-Tier Parser Daemons

Native daemon → tree-sitter → regex fallback

Node.js Daemon (TS/JSX)Rust Daemon (Rust/C/C++)JVM Daemon (Java)tree-sitter (fallback)

LiquidGraph In-Memory DB

6-index graph with persistent storage

node_id → GraphNodefile_path → {node_ids}source_id → [edges]JSON persistence

Incremental Pipeline (0.05ms total)

Watch data flow through each stage in real-time

File Edit
0.00ms
AST Extract
0.02ms
Delta Compute
0.01ms
Graph Patch
0.01ms
Edge Resolve
0.01ms
Propagate
0.00ms
Total: 0.05msper incremental update

Relationship Types

calls
fn() → fn()
imports
import → def
inherits
class → class
uses_type
var: Type
decorated
@dec → fn
Real Benchmarks

Numbers that speak for themselves

Tested on a real 18K LOC production app (TypeScript + Rust, Tauri 2) with Claude Sonnet 4.5 across 10 diverse tasks.

Metric
Without
With StreamRAG
Delta
Accuracy
73%
78%
+5pp
Total Time
966s
821s
15% ↓
Total Cost
$6.64
$5.87
12% ↓
Perfect Scores
3/10
5/10
+2
+25pp
75% → 100% accuracy
Cross-Module Understanding
52%
fewer tokens consumed
Dependency Analysis
33%
faster completion
Architecture Tracing
11.6%
reduced total cost
Cost Savings

Benchmark: Production App (18K LOC) · Model: Claude Sonnet 4.5 · 10 tasks across 4 categories

Multi-Language

7 Languages, three-tier extraction

Native daemon for speed, tree-sitter for accuracy, regex for universal coverage. Automatic fallback — always the best parser available.

Tier 1
Native Daemons
Persistent processes using language-native compilers. Fastest, most accurate.
Tier 2
tree-sitter
Incremental parsing library. Great for broken mid-edit code recovery.
Tier 3
Regex Fallback
Pattern-based extraction. Universal coverage, zero dependencies.
Py
Python
.py
stdlib ast
TS
TypeScript
.ts .tsx
Node.js Daemontree-sitterregex
Rx
React JSX
.jsx .tsx
Node.js Daemontree-sitterregex
JS
JavaScript
.js .mjs .cjs
Node.js Daemonregex
Rs
Rust
.rs
Rust Daemontree-sitterregex
C+
C / C++
.c .cpp .h
Rust Daemonregex
Jv
Java
.java
JVM Daemonregex
Get Started

Install in seconds

Zero configuration required. StreamRAG activates automatically on your first edit and builds the graph in the background.

terminal
claude plugin marketplace add Krrish2004/StreamRAG

One command. Installs from the official Claude Code plugin marketplace.

Python
3.9+
License
Proprietary
Size
20K+ LOC

Ready to give your AI superpowers?

One command. Zero config. StreamRAG starts building your code graph the moment you begin editing. Join the developers who stopped burning tokens on exploration.

20K+Lines of Code
1700+Tests Passing
98Test Files