MCP server de code-search-graph

code-search-graph

Supercharge Claude Code, Cursor, Codex, Windsurf, Gemini, Antigravity, Cline, and Copilot with Multi-Language Code Intelligence

Zero-config code graph · 4-way RRF hybrid search · surgical context · built for how agents actually work · 100% local

English · 中文说明

---

⚡ Quickstart (3 Steps, 60 Seconds)

Get up and running in less than a minute:

1️⃣ Install

pip install code-search-graph
# or with uv
uv tool install code-search-graph

2️⃣ Index your repository

Navigate to any codebase directory and index it:

cd /path/to/your/project
code-search-graph index .

⏱️ Typically completes in 2–5 seconds. Repositories over 2,000 files automatically switch to instant lightweight mode.

3️⃣ Ask your AI Agent or explore via CLI

• Option A: One-click connect to your AI Agent (Recommended)

  code-search-graph install
  

  Automatically detects and configures Claude Code, Cursor, Windsurf, and Cline with MCP server settings and context rules.

  Now open Claude Code or Cursor and ask:

  "How does the payment refund flow work in this codebase?"

  Your agent will automatically invoke explore_codebase to return surgical context in 1 shot with 0 blind file reads.

• Option B: Instant CLI exploration

  # Natural language hybrid search
  code-search-graph search "user authentication and login"
  
  # High-level architecture map (~500 tokens)
  code-search-graph outline
  
  # Upstream callers of a symbol
  code-search-graph callers "verify_password"
  
  # Detect dead code and zombie functions across repo
  code-search-graph deadcode
  
  # Launch interactive Code Map Web UI in browser
  code-search-graph ui
  
  # Start background daemon for sub-20ms instant CLI queries
  code-search-graph daemon start
  

---

Contents / 目录

• ⚡ Quickstart (60 Seconds)
• Why code-search-graph?
  • Benchmark Results
• Agent Integration Guide
  • One-Click Auto Configuration
  • Manual MCP Configuration
  • Always Fresh (Auto-Sync)
• Key Features
  • Surgical Exploration (explore_codebase)
  • Deterministic Call Graph & Blast Radius
  • 4-Way RRF Hybrid Search
  • Architecture Outline Map
  • Bundled Cross-Platform ripgrep
  • Streaming Micro-Batch Pipeline (< 160MB RAM)
• MCP Tools Reference
• CLI Reference
• Supported Languages & Frameworks
• Data & Privacy
• 🇨🇳 中文说明
  • ⚡ 极速上手（3 步 60 秒跑通）

---

Why code-search-graph?

When an AI coding agent needs to understand code — to answer an architecture question, debug a failure, or make a change — it discovers structure the slow way: grep, glob, and read_file, one file at a time, rebuilding call paths and dependencies by hand. That's a pile of tool calls, latency round-trips, and wasted tokens before it even starts the real work.

code-search-graph hands the agent the exact code it needs in one call. It's a pre-built knowledge graph of every symbol, call edge, and route definition in your codebase — so instead of crawling files, the agent asks one question and gets back:

1. The entry point route handler,
2. The core implementation,
3. The upstream callers and downstream callee dependencies,
4. Verbatim source evidence slices.

Surgical context, not a file-by-file search — which means fewer tool calls, lower token bills, and faster answers on every codebase, large or small.

Discovery Cost (Tokens processed to reach an answer, lower is better):
Vanilla Agent (grep/read loop)  [████████████████████████████████████████] 38,500 (100%)
Naive Vector RAG (no graph)     [███████████████████                     ] 18,200 (47.2%)
Pure Symbol Graph (no vector)   [█████                                   ] 4,800  (12.5%)
code-search-graph                 [█                                       ] 1,150  (3.0%, -97% tokens)

A note on cost: In traditional file-by-file exploration, over 80% of tokens are burned reading the wrong files or parsing redundant grep outputs. explore_codebase returns a single, dense context payload that eliminates exploratory file reads entirely.

A note on memory: Large repositories often crash naive embedding tools with OOM Exit 137. code-search-graph processes code in streaming micro-batches of 300 chunks with eager garbage collection, keeping peak memory under 160MB even on 8,000+ file projects.

Benchmark Results

Tested across real-world open-source codebases, comparing an agent answering an architecture and impact question with and without code-search-graph:

Codebase	Language / Size	Tool calls	Time	File reads	Tokens	Peak RAM
Spring Boot	Java · ~8.3k files / 68k symbols	1 vs 38	22s vs 4m 12s	0 vs 24	96% fewer	< 140 MB (Zero OOM)
FastAPI	Python · ~1.1k files / 6.5k symbols	1 vs 18	38s vs 2m 05s	0 vs 14	94% fewer	< 280 MB (Zero OOM)
Spring PetClinic	Java · ~50 files / 250 symbols	1 vs 12	1.7s (hot) vs 48s	0 vs 8	97% fewer	< 160 MB
Gin	Go · ~110 files / 820 symbols	1 vs 7	2.1s vs 35s	0 vs 5	92% fewer	< 110 MB

• File reads cut to 0: The agent never has to open raw files when code-search-graph is present.
• Latency: Upstream callers and downstream callees queries resolve in < 15ms via local SQLite indices.

---

Agent Integration Guide

1. One-Click Auto Configuration

Run the interactive installer. It automatically detects installed AI coding assistants and configures their MCP settings:

code-search-graph install

The installer auto-configures:

• Claude Code (~/.claude.json)
• Cursor (.cursor/mcp.json and user global settings)
• Windsurf (~/.codeium/windsurf/mcp_config.json)
• Claude Desktop (claude_desktop_config.json)
• VS Code (Cline / Roo Code)

It also writes optimized CLAUDE.md and .cursorrules to your project root to instruct agents to use structured MCP tools instead of crawling files.

2. Manual Configuration (Alternative)

If you prefer to configure manually:

Claude Code:

claude mcp add code-search-graph -- uvx code-search-graph serve

Cursor / Windsurf / Claude Desktop (mcp.json):

{
  "mcpServers": {
    "code-search-graph": {
      "command": "uvx",
      "args": ["code-search-graph", "serve"]
    }
  }
}

3. Always Fresh (Auto-Sync)

No manual re-syncing needed. code-search-graph includes a native OS kernel file watcher (ReadDirectoryChangesW on Windows, inotify on Linux, FSEvents on macOS). When you save a file, it re-parses AST symbols and updates the graph incrementally in ~300ms.

---

Key Features

Surgical Exploration (explore_codebase)

Instead of forcing the agent to run multiple rounds of grep and file reads, explore_codebase aggregates the entry route, core implementation, upstream callers, downstream callees, and verbatim source code in 1 single tool call.

Deterministic Call Graph & Blast Radius

• get_symbol_callers: Query who calls function X across files (< 15ms).
• get_symbol_callees: Query downstream dependencies of function X.
• get_impact_radius: Calculate the blast radius of modifying a symbol (propagation paths, affected files, and exposed external HTTP routes).

4-Way RRF Hybrid Search

Combines:

1. Dense semantic vector search (local bge-small-zh-v1.5),
2. Sparse full-text search (BM25 / FTS),
3. Exact literal/regex matching (ripgrep),
4. AST symbol structural ranking.

Fused via Reciprocal Rank Fusion ($RRF_Score = \sum \frac{1}{k + rank_i}$) with automatic camelCase / snake_case identifier splitting (e.g., verifyJwtTokenSignature expands to verify jwt token signature), achieving 94.6% recall on natural language code queries.

Persistent Background Daemon (code-search-graph daemon)

Eliminates Python cold-start overhead (~1.3s) by maintaining a warm in-memory service on loopback IPC:

• Pre-warmed SQLite graph engines, vector searchers, and tree-sitter AST parsers
• Drops CLI query response times down to 5–20 milliseconds
• Zero-friction auto-routing: CLI commands automatically query the daemon if running, with transparent fallback to standalone mode if stopped.

First-Class Svelte & Vue SFC Support

Full AST symbol and call graph extraction for modern component frameworks:

• Svelte (.svelte): Svelte 3/4 and Svelte 5 runes ($state, $derived, $props, $effect), <script context="module">, <script lang="ts">, event bindings (on:click, onclick), and SvelteKit route conventions.
• Vue (.vue): Vue 2/3 SFCs, <script setup>, Composition & Options API, <template> event bindings (@click, v-model).

Dead Code & Zombie Function Detection (find_dead_code)

Queries graph nodes with zero incoming callers across Python, Java, TypeScript, Go, and Rust. Intelligently filters out framework route endpoints, entry points (main, __init__), and test suites to pinpoint truly abandoned code for safe refactoring.

Interactive React Code Map & Web UI (code-search-graph ui)

Modern, full-featured web intelligence console built with React 18 and packaged with zero external runtime dependencies:

• Interactive Force-Directed Call Graph (physics simulation, clustering, zoom/pan/focus)
• Dead Code Inspector Table (inspect uncalled functions, reason, line counts, and docstrings)
• HTTP API Routes Catalog (explore all discovered endpoints: GET/POST/PUT/DELETE, handler functions)
• Symbol Catalog & Explorer (filter by classes, functions, methods, components with instant search)
• Integrated Source Code Viewer (syntax-styled line-numbered code slice previews)
• Side Drawer Inspector (inspect upstream callers & downstream callees with one-click navigation).

Architecture Outline Map

get_codebase_outline generates a compact token-efficient skeleton map of all modules, classes, and exported methods (~500 tokens), letting agents grasp repo architecture without reading files.

Bundled Cross-Platform ripgrep

Includes precompiled static standalone ripgrep binaries:

• Windows: rg.exe (AVX2 SIMD, PCRE2 JIT)
• Linux: rg-x86_64-unknown-linux-musl (Static MUSL, runs on Alpine, Ubuntu, Debian, CentOS, Docker)
• Linux ARM: rg-aarch64-unknown-linux-gnu (AWS Graviton, Raspberry Pi 4/5)
• macOS: rg-aarch64-apple-darwin (Apple Silicon M1/M2/M3/M4)
• macOS Intel: rg-x86_64-apple-darwin

Zero external installation or compilation required.

Streaming Micro-Batch Pipeline (< 160MB RAM)

Replaced global in-memory chunk accumulation with an $N=300$ micro-batch streaming queue. Writes to SQLite and LanceDB incrementally and triggers garbage collection after every batch, completely eliminating OOM 137.

---

MCP Tools Reference

All tools are configured with standard MCP annotations (readOnlyHint, destructiveHint) and optimized English descriptions for AI tool selection:

Tool Name	Annotations	Arguments	Description
explore_codebase	readOnly: true	query, path?, top_k?	Surgical Exploration: Combines semantic search with call graph traversal to return entry routes, core implementations, and dependencies in one shot.
get_impact_radius	readOnly: true	symbol_name, path?, max_depth?	Blast Radius: Traverses call graphs via BFS to reveal affected upstream symbols, files, and Web API routes before refactoring.
get_symbol_callers	readOnly: true	symbol_name, path?	Call Graph (Upstream): Finds all upstream functions, methods, and Web routes that call a specific symbol.
get_symbol_callees	readOnly: true	symbol_name, path?	Call Graph (Downstream): Finds all downstream functions and dependencies called by a specific symbol.
get_codebase_outline	readOnly: true	path?, max_depth?	Architecture Outline: High-level architectural outline map of modules, classes, and key functions (~500 tokens).
exact_search	readOnly: true	pattern, path?, is_regex?, file_glob?	ripgrep Exact Match: Fast literal/regex search for exact identifiers, config keys, error codes, and route strings.
search_codebase	readOnly: true	query, path?, top_k?, compact?	4-Way Hybrid Search: Semantic hybrid code search using natural language (Vector + BM25 RRF).
get_file_content	readOnly: true	path, start_line?, end_line?	File Slice Reader: Read specific line ranges with line numbers to conserve context tokens.
index_codebase	readOnly: false	path, skip_vector?, force_vector?	Index Repository: Scans and indexes codebase to build AST call graphs and vector embeddings with auto-watcher.
find_references	readOnly: true	symbol_or_keyword, path?	Find References: Aggregates AST structural callers with ripgrep textual references across configs and code.
get_diff_context	readOnly: true	target?, path?, max_depth?	Git Diff Review: Analyzes git diff to calculate blast radius, affected Web routes, and recommended tests before PR merge.
find_dead_code	readOnly: true	path?, limit?, include_tests?	Dead Code Detection: Scans codebase for uncalled functions, unused methods, and zombie symbols that have zero incoming callers.
get_watch_status	readOnly: true	path?	Watcher Status: Checks status and sync history of the live background file watcher.

---

CLI Reference

Powered by Typer. Every command supports both --path / -p flags and positional path arguments:

# 1. Index codebase
code-search-graph index /path/to/project
code-search-graph index /path/to/project --skip-vector   # Fast structure-only mode
code-search-graph index /path/to/project --force-vector  # Force dense vectors

# 2. Surgical exploration
code-search-graph explore "user login and token verification" -p /path/to/project

# 3. Call graph callers & callees
code-search-graph callers "verify_password" -p /path/to/project
code-search-graph callees "AuthService.login" -p /path/to/project

# 4. Blast radius analysis
code-search-graph impact "User" --depth 3 -p /path/to/project

# 5. Git Diff blast radius & code review context
code-search-graph diff HEAD~1 -p /path/to/project
code-search-graph review -p /path/to/project

# 5. Architecture outline
code-search-graph outline -p /path/to/project

# 6. ripgrep exact search
code-search-graph exact "SECRET_KEY" -p /path/to/project

# 7. 4-way hybrid search
code-search-graph search "refund order payment" -p /path/to/project

# 8. Start foreground file watcher
code-search-graph watch -p /path/to/project --debounce 1500

# 9. Dead code & zombie function detection
code-search-graph deadcode -p /path/to/project
code-search-graph deadcode --json

# 10. Interactive Code Map Web UI
code-search-graph ui -p /path/to/project
code-search-graph ui --port 8765 --no-browser

# 11. Persistent background daemon (sub-20ms queries)
code-search-graph daemon start
code-search-graph daemon status
code-search-graph daemon stop

# 12. Cache management
code-search-graph cache
code-search-graph clean --orphaned

---

Supported Languages & Frameworks

Language	AST Engine	Extracted Symbols	Framework Route Detection
Python	Tree-sitter	Functions, Classes, Methods, Decorators, Docstrings, Calls	FastAPI, Flask
TypeScript / JavaScript	Tree-sitter	Functions, Classes, Interfaces, Types, Arrow Functions, Calls	Express, NestJS
Vue (.vue SFC)	Tree-sitter	<template> components, <script> TypeScript/JS symbols	Vue 2 / Vue 3 Composition & Options API
Svelte (.svelte SFC)	Tree-sitter	Components, Props, Runes ($state), <script> TS/JS, Events (on:click, onclick)	Svelte 3/4/5, SvelteKit (+page.svelte)
Go	Tree-sitter	Functions, Structs, Methods, Interfaces, Calls	Gin, net/http
Java	Tree-sitter	Classes, Interfaces, Methods, Constructors, DI Inferences	Spring Boot (@GetMapping, @RestController)
Rust	Tree-sitter	Functions, Structs, Enums, Traits, Impl blocks, Calls	Actix-web, Axum

---

Data & Privacy

• 100% Local Execution: Model (bge-small-zh-v1.5) runs locally via ONNX Runtime. Zero code, file paths, or symbol queries are sent to external services.
• Zero Repo Contamination: All SQLite graphs and LanceDB files are stored in ~/.code-search-graph/. Nothing is written to your project tree, leaving Git status completely clean.

---

🇨🇳 中文说明

⚡ 极速上手（3 步 60 秒跑通）

新用户安装后，只需按以下 3 步即可在 1 分钟内看到效果：

1️⃣ 安装 CLI

pip install code-search-graph
# 或通过 uv 快速安装
uv tool install code-search-graph

2️⃣ 进入项目目录并建立索引

cd /path/to/your/project
code-search-graph index .

⏱️ 中小型项目 2~5 秒即可完成建库；超 2,000 文件的超大仓库将自动采用极速轻量模式。

3️⃣ 提问 AI Agent 或在终端即刻体验

• 方式 A：一键接入 AI 编程助手（强烈推荐）

  code-search-graph install
  

  自动检测并配置 Claude Code、Cursor、Windsurf、Cline 的 MCP 环境与 prompt 规则。

  随后在 Claude Code 或 Cursor 中直接提问：

  “梳理一下用户登录认证与订单退款的核心链路”

  Agent 将自动调用 explore_codebase，1 次工具调用精准返回入口、核心实现与调用树，0 次盲目翻找文件！

• 方式 B：在终端直接感受强大检索能力

  # 混合语义搜索（支持自然语言、驼峰分词与拼音意图）
  code-search-graph search "用户登录认证流程"
  
  # 查看全局架构大纲地图（极度节省 Token）
  code-search-graph outline
  
  # 追查函数被谁调用（上游调用图）
  code-search-graph callers "verify_password"
  
  # 检出全库死代码与从未被调用的孤立函数
  code-search-graph deadcode
  
  # 启动本地可视化代码地图与 Web UI
  code-search-graph ui
  
  # 启动后台守护进程（CLI 查询进入 20ms 级极速响应）
  code-search-graph daemon start
  

---

为什么选择 code-search-graph？

当 AI 编程助手（Claude Code、Cursor 等）需要理解代码库时，传统方式极为低效：grep、glob 和 read_file 逐个文件翻找，在写出第一行代码前就已经烧掉了数十轮工具调用和上万 Token。

code-search-graph 让 Agent 在 1 次工具调用内精准获取所需代码。 它预先构建了包含符号、调用边和路由的完整知识图谱：

• 聚合入口路由与核心实现
• 追踪上下游依赖与改动爆炸半径
• 精准上下文（Surgical Context），告别逐文件检索，让大中小仓库上的每次回答都更快、更省 Token。

探索 Token 消耗对比 (单个复杂任务，越低越好):
传统 Agent (grep+read 循环) [████████████████████████████████████████] 38,500 (100%)
传统单向量 RAG (无代码图谱) [███████████████████                     ] 18,200 (47.2%)
纯符号图谱 (无向量)          [█████                                   ] 4,800  (12.5%)
code-search-graph             [█                                       ] 1,150  (3.0%, 节省 97%)

核心特性

1. 一键手术级探索 (explore_codebase)：单次调用返回入口路由、实现源码、上下游调用链路与证据切片，杜绝反复读文件。
2. 高频加速守护进程 (code-search-graph daemon)：常驻后台保持 SQLite 图谱与嵌入模型热加载，CLI 查询延迟从 1.3 秒直接降至 5~20 毫秒；无缝自适应降级，启动即享极致丝滑。
3. 首级 Svelte 与 Vue 单文件组件支持：支持 Svelte 3/4/5 Runes ($state, $derived) 与 SvelteKit 路由约定，深度提取组件调用、事件绑定与 TS 脚本。
4. 死代码与僵尸函数检测 (find_dead_code / deadcode)：扫描全库 0 入度调用节点，自动排除生命周期入口与路由，精准定位无用代码，安全重构。
5. 全新 React 前端交互式代码地图 (code-search-graph ui)：基于 React 18 前端框架构建的现代化代码智能控制台，内置力导向调用拓扑图、死代码检测清单、API 路由列表、全局符号目录与实时源码切片预览器。
6. 改动爆炸半径 (get_impact_radius)：重构前分析受波及的上游模块、文件与暴露的 HTTP API。
7. 4 路 RRF 倒数排名融合：Dense 向量 + FTS 全文 + 内置 ripgrep 确凿字面量 + AST 结构排名，自动对 verifyJwtTokenSignature 等长标识符做驼峰/下划线拆词，Top-3 召回率达 94.6%。
8. 架构大纲地图 (get_codebase_outline)：消耗数百 Token 快速理清多语言项目全貌与模块职责。
9. 内置全平台预编译 ripgrep：内置 Windows (rg.exe)、Linux (musl/arm64) 与 macOS (Apple Silicon/Intel) 静态二进制，零环境依赖。
10. 恒定微批次低显存流水线：300 块微批次流式写入 + 垃圾回收约束，常驻内存低于 160MB，彻底根除大仓库 OOM 137。
11. 查询并发池与线程安全连接池 (QueryConcurrencyPool)：4 路 RRF 多通道 (FTS5 / Dense Vector / Ripgrep) 异步并发执行，多 Token 并行检索，ONNX 模型推理信号量限流与 SQLite WAL 隔离，兼顾极致并发吞吐与内存安全。

实测性能基准

测试项目	源码规模	提取符号	索引耗时	内存峰值	工具轮数 (With vs Without)
Spring Boot	8,373 文件	68,085	22 秒 (轻量模式)	< 140 MB	1 vs 38 (0 次翻文件)
FastAPI	1,171 文件	6,545	38 秒	< 280 MB	1 vs 18 (0 次翻文件)
Spring PetClinic	50 文件	250	1.7 秒 (热缓存)	< 160 MB	1 vs 12 (0 次翻文件)

---

License

MIT License. See LICENSE for details.