depwire
Code dependency graph and AI context engine. 10 MCP tools that give Claude, Cursor, and any MCP client full codebase context — impact analysis, dependency tracing, architecture summaries, and interactive arc diagram visualization. Supports TypeScript, JavaScript, Python, and Go.
README
Depwire

See how your code connects. Give AI tools full codebase context.
Depwire analyzes codebases to build a cross-reference graph showing how every file, function, and import connects. It provides:
- 🎨 Beautiful arc diagram visualization — Interactive Harrison Bible-style graphic
- 🤖 MCP server for AI tools — Cursor, Claude Desktop get full dependency context
- 🔍 Impact analysis — "What breaks if I rename this function?" answered precisely
- 👀 Live updates — Graph stays current as you edit code
- 🌍 Multi-language — TypeScript, JavaScript, Python, and Go
Why Depwire?
AI coding tools are flying blind. Every time Claude, Cursor, or Copilot touches your code, it's guessing about dependencies, imports, and impact. The result: broken refactors, hallucinated imports, and wasted tokens re-scanning files it already saw.
Lost context = lost money + lost time + bad code.
Depwire parsed the entire Hono framework — 305 files, 5,636 symbols, 1,565 dependency edges — in 2.3 seconds.
Depwire fixes this by giving AI tools a complete dependency graph of your codebase — not a fuzzy embedding, not a keyword search, but a deterministic, tree-sitter-parsed map of every symbol and connection.
Stop Losing Context
- No more "start from scratch" chats — Depwire is the shared knowledge layer that every AI session inherits. New chat? Your AI already knows the architecture.
- Stop burning tokens — AI tools query the graph instead of scanning hundreds of files blindly
- One command, every AI tool — Claude Desktop, Cursor, VS Code, any MCP-compatible tool gets the same complete picture
Ship Better Code
- Impact analysis for any change — renaming a function, moving a file, upgrading a dependency, deleting a module — know the full blast radius before you touch anything
- Refactor with confidence — see every downstream consumer, every transitive dependency, 2-3 levels deep
- Catch dead code — find symbols nobody references anymore
Stay in Flow
- Live graph, always current — edit a file and the dependency map updates in real-time. No re-indexing, no waiting.
- Works locally, stays private — zero cloud accounts, zero data leaving your machine. Just
npm installand go.
10 MCP Tools, Not Just Visualization
Depwire isn't just a pretty graph. It's a full context engine with 10 tools that AI assistants call autonomously — architecture summaries, dependency tracing, symbol search, file context, and more. The AI decides which tool to use based on your question.
Installation

npm install -g depwire-cli
Or use directly with npx:
npx depwire-cli --help
Quick Start
CLI Usage
# Visualization (opens in browser)
npx depwire-cli viz ./my-project
# Parse and export as JSON
npx depwire-cli parse ./my-project
# Exclude test files and node_modules
npx depwire-cli parse ./my-project --exclude "**/*.test.*" "**/node_modules/**"
# Show detailed parsing progress
npx depwire-cli parse ./my-project --verbose
# Export with pretty-printed JSON and statistics
npx depwire-cli parse ./my-project --pretty --stats
# Custom output file
npx depwire-cli parse ./my-project -o my-graph.json
Claude Desktop
Add to your Claude Desktop config (~/Library/Application Support/Claude/claude_desktop_config.json on macOS):
{
"mcpServers": {
"depwire": {
"command": "npx",
"args": ["-y", "depwire-cli", "mcp"]
}
}
}
Then in chat:
Connect to /path/to/my/project and show me the architecture.
Cursor
Settings → Features → Experimental → Enable MCP → Add Server:
- Command:
npx - Args:
-y depwire-cli mcp /path/to/project
Available MCP Tools
| Tool | What It Does |
|---|---|
connect_repo |
Connect to any local project or GitHub repo |
impact_analysis |
What breaks if you change a symbol? |
get_file_context |
Full context — imports, exports, dependents |
get_dependencies |
What does a symbol depend on? |
get_dependents |
What depends on this symbol? |
search_symbols |
Find symbols by name |
get_architecture_summary |
High-level project overview |
list_files |
List all files with stats |
get_symbol_info |
Look up any symbol's details |
visualize_graph |
Generate interactive arc diagram visualization |
Supported Languages
| Language | Extensions | Features |
|---|---|---|
| TypeScript | .ts, .tsx |
Full support — imports, classes, interfaces, types |
| JavaScript | .js, .jsx, .mjs, .cjs |
ES modules, CommonJS require(), JSX components |
| Python | .py |
Imports, classes, decorators, inheritance |
| Go | .go |
go.mod resolution, structs, interfaces, methods |
Visualization


# Open visualization on default port (3456)
depwire viz ./my-project
# Custom port
depwire viz ./my-project --port 8080
# Exclude test files from visualization
depwire viz ./my-project --exclude "**/*.test.*"
# Verbose mode with detailed parsing logs
depwire viz ./my-project --verbose
# Don't auto-open browser
depwire viz ./my-project --no-open
Opens an interactive arc diagram in your browser:
- Rainbow-colored arcs showing cross-file dependencies
- Hover to explore connections
- Click to filter by file
- Search by filename
- Live refresh when files change — Edit code and see the graph update in real-time
- Export as SVG or PNG
- Port collision handling — Automatically finds an available port if default is in use
How It Works
- Parser — tree-sitter extracts every symbol and reference
- Graph — graphology builds an in-memory dependency graph
- MCP — AI tools query the graph for context-aware answers
- Viz — D3.js renders the graph as an interactive arc diagram
CLI Reference
depwire parse <directory>
Parse a project and export the dependency graph as JSON.
Options:
-o, --output <path>— Output file path (default:depwire-output.json)--exclude <patterns...>— Glob patterns to exclude (e.g.,"**/*.test.*" "dist/**")--verbose— Show detailed parsing progress (logs each file as it's parsed)--pretty— Pretty-print JSON output with indentation--stats— Print summary statistics (file count, symbol count, edges, timing)
Examples:
# Basic parse
depwire parse ./src
# Exclude test files and build outputs
depwire parse ./src --exclude "**/*.test.*" "**/*.spec.*" "dist/**" "build/**"
# Full verbosity with stats
depwire parse ./src --verbose --stats --pretty -o graph.json
depwire viz <directory>
Start visualization server and open arc diagram in browser.
Options:
--port <number>— Port number (default: 3456, auto-increments if in use)--exclude <patterns...>— Glob patterns to exclude--verbose— Show detailed parsing progress--no-open— Don't automatically open browser
Examples:
# Basic visualization
depwire viz ./src
# Custom port without auto-open
depwire viz ./src --port 8080 --no-open
# Exclude test files with verbose logging
depwire viz ./src --exclude "**/*.test.*" --verbose
depwire mcp [directory]
Start MCP server for AI tool integration (Cursor, Claude Desktop).
Examples:
# Start MCP server on current directory
depwire mcp
# Start on specific project
depwire mcp /path/to/project
Error Handling
Depwire gracefully handles parse errors:
- Malformed files — Skipped with warning, parsing continues
- Large files — Files over 1MB are automatically skipped
- Port collisions — Auto-increments to next available port (3456 → 3457 → 3458...)
- Protected paths — Blocks access to sensitive directories (.ssh, .aws, /etc)
Example Workflows
Refactoring with AI

# In Claude Desktop or Cursor with Depwire MCP:
"Connect to /Users/me/my-app and analyze the impact of renaming UserService to UserRepository"
# Depwire responds with:
# - All files that import UserService
# - All call sites
# - All type references
# - Suggested find-and-replace strategy
Understanding a New Codebase
"Connect to https://github.com/t3-oss/create-t3-app and give me an architecture summary"
# Depwire responds with:
# - Language breakdown
# - Module/package structure
# - Most-connected files (architectural hubs)
# - Entry points
Pre-Commit Impact Check
# Check what your changes affect before committing
depwire viz . --open
# Review the arc diagram — red arcs show files you touched
Security
Depwire is read-only — it never writes to, modifies, or executes your code.
- Parses source files with tree-sitter (the same parser used by VS Code and Zed)
- Visualization server binds to localhost only
- No data leaves your machine — everything runs locally
- Blocks access to sensitive system directories (.ssh, .aws, /etc)
- npm packages published with provenance verification
See SECURITY.md for full details.
Roadmap
- [ ] PR Impact Visualization (GitHub Action)
- [ ] Temporal Graph — watch your architecture evolve over git history
- [ ] Cross-language edge detection (API routes ↔ frontend calls)
- [ ] Dependency health scoring
- [ ] VSCode extension
Contributing
Contributions welcome! Please note:
- Fork the repository
- Create a feature branch
- Add tests for new functionality
- Submit a pull request
- Sign the CLA (handled automatically on your first PR)
All contributors must sign the Contributor License Agreement before their PR can be merged.
Author
Atef Ataya — AI architect, author, and creator of Depwire.
- 🎥 YouTube — 600K+ subscribers covering AI agents, MCP, and LLMs
- 📖 The Architect's Playbook: 5 Pillars — Best practices for AI agent architecture
License
Depwire is licensed under the Business Source License 1.1.
- Use it freely for personal projects, internal company use, and development
- Cannot be offered as a hosted/managed service to third parties
- Converts to Apache 2.0 on February 25, 2029
For commercial licensing inquiries: atef@depwire.dev
Credits
Built by ATEF ATAYA LLC
Powered by:
- tree-sitter — Fast, reliable parsing
- graphology — Powerful graph data structure
- D3.js — Data visualization
- Model Context Protocol — AI tool integration
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。