codesynapse

Give your AI assistant a map of your codebase

Architecture-level code intelligence delivered over MCP. Class hierarchies, call chains, blast radius — answered from a graph, not file search.

Get started View on GitHub
curl -fsSL https://raw.githubusercontent.com/sohilladhani/codesynapse/master/install.sh | sh
32 MCP tools
30+ languages
100% local
0 cloud APIs
codesynapse demo — module add, module list, query

Hybrid BM25 + dense search

BM25 for exact symbol matches, dense vector embeddings for semantic concepts. RRF fusion means "auth token handler" finds the right class — even if it's called JwtValidator.

Blast radius analysis

Know the full impact of a change before touching a line. Trace callers, transitive dependents, and cross-module ripple effects — scored by risk.

Fully local, no cloud

Runs on CPU. No GPU, no API keys, no data leaving your machine. The embedding model downloads once — every query after that is instant and free.

Works with

  • Claude Code
  • Cursor
  • OpenCode
  • Codex CLI
  • Hermes Agent
  • Kiro
  • any MCP client

Quick start

  1. 1

    Install

    Pre-built binary — no Rust required.

    curl -fsSL https://raw.githubusercontent.com/sohilladhani/codesynapse/master/install.sh | sh

    Rust user? cargo install codesynapse-cli · or grab a pre-built binary.

  2. 2

    Register with your AI client

    Auto-detects Claude Code and Cursor. For other clients, see flags below.

    codesynapse setup

    Other clients: codesynapse opencode install · codesynapse codex install · codesynapse kiro install

  3. 3

    Index a repository

    codesynapse module add myrepo /path/to/repo

    Add multiple repos — graphs merge automatically for cross-module queries.

  4. 4

    Ask architecture questions

    Restart your AI client. The 32 MCP tools are now available.

    "What handles auth token expiry?" "Show blast radius of UserService" "Trace callers of PaymentProcessor.charge"

How it compares

codegraph semble Sourcegraph Cody codesynapse
MCP tools 10 2 — 32
Hybrid BM25 + dense ✗ ✓ ✓ ✓
Structural graph ✓ ✗ ✓ ✓
Blast radius ✓ ✗ ✓ ✓
Fully local ✓ ✓ ✗ ✓
No cloud API needed ✓ ✓ ✗ ✓
Multi-module graph ✓ ✗ ✓ ✓
No telemetry by default ✗ opt-out — ✗ ✓ opt-in

codegraph is the closest alternative — TypeScript, local, blast radius, multi-module. Key gap: FTS5 lexical search only (no dense embeddings), so concept queries miss results BM25+dense RRF catches. semble uses the same Model2Vec + BM25 + RRF stack but is search-only — no structural graph. Sourcegraph Cody has graph analysis but requires cloud infrastructure.