Understand-Anything MCP Server
Bridges codebase structural analysis, architectural rules, and CI/CD validation with MCP clients, enabling LLMs to check blast radius and prevent breaking changes.
README
Understand-Anything MCP Server
This MCP Server bridges Egonex-AI/Understand-Anything with Claude Desktop and other MCP clients, giving LLMs native access to full-codebase structural graphs, architecture analysis, and CI/CD validation to prevent breaking changes.
Privacy Policy
[!IMPORTANT] Link to Privacy Policy
Data Processing Details:
- License keys and email addresses are securely stored for billing purposes.
- Purely local, no network calls:
ua_status,ua_scan,ua_graph_summary,ua_explain,ua_onboarding_doc.- Sends graph data to the backend (on both Free and Pro):
ua_precheck,ua_find_callers,ua_impact_analysis,ua_rules,ua_ci_check,ua_validate_graph. When these tools are used, the full local graph object is sent per request to our backend for processing, license, and quota validation.- No source code contents are transmitted, only graph metadata (file paths and import relationships). All backend graph processing is done purely in-memory per-request and is never persisted.
Features
Most code intelligence tools tell your agent what's connected. ua-mcp tells your agent what it's allowed to change — and flags risky changes before they happen.
- Advanced Context Analysis: Export targeted contexts or the full graph for advanced LLM reasoning.
- Open-source with Premium Tiers: Free essential features, with advanced capabilities for Pro users.
[!TIP] System Instruction Recommended: For the best experience, add the following to your AI assistant's system prompt or custom instructions: "Always call
ua_precheckbefore modifying any file in this project."
How it works
- Your AI agent decides it wants to modify a critical file (e.g.,
src/auth.ts). - The agent (following its system instructions) triggers
ua_precheckbefore making the edit to run an Architectural Safety Linter. - The server analyzes the graph to determine the "blast radius" and checks it against your rules.
- If the blast radius is too large or violates a rule, a Safety Checkpoint (Elicitation Prompt) interrupts the agent, asking for your explicit confirmation before proceeding.
Configuration Rules (.ua-rules.json)
You can define specific boundaries in a .ua-rules.json file in the root of your workspace to dictate what the LLM is allowed to touch.
{
// Understand-Anything Architectural Rules
// Define constraints that agents and developers must respect.
"rules": [
{
"id": "no-ui-db-import",
"description": "UI layer must never import database layer directly",
"from_pattern": "src/ui/**",
"to_pattern": "src/db/**",
"severity": "error"
},
{
"id": "auth-required-for-payments",
"description": "Payment modules must always be reachable from auth",
"requires_path_through": "src/auth/**",
"for_pattern": "src/payments/**",
"severity": "error"
}
]
}
Quick Start
Install the server globally:
npm install -g ua-mcp
Client Configuration
Add the following to your MCP client configuration file:
Claude Desktop (claude_desktop_config.json)
Add the following to your Claude Desktop config file (usually ~/Library/Application Support/Claude/claude_desktop_config.json on macOS or %APPDATA%\Claude\claude_desktop_config.json on Windows):
{
"mcpServers": {
"understand-anything": {
"command": "npx",
"args": ["-y", "ua-mcp"],
"env": {
"UA_PROJECT_PATH": "/path/to/your/project",
"UA_LICENSE_KEY": "your_license_key_here"
}
}
}
}
[!WARNING] Claude Desktop Bug (July 2026): Claude Desktop currently has a bug where MCP elicitation requests (form prompts) silently auto-cancel instead of showing the dialog UI. If your agent says the precheck was cancelled without you seeing a prompt, this is a Claude Desktop client bug, not a server failure.
Cursor
- Go to Settings > Features > MCP.
- Click + Add new MCP server.
- Name:
understand-anything - Type:
command - Command:
npx -y ua-mcp
Continue
Add to your config.json under mcpServers:
"understand-anything": {
"command": "npx",
"args": ["-y", "ua-mcp"]
}
Available Tools & Tiering
The Understand-Anything MCP Server operates on a tiered licensing model.
Core Tools (Free Tier)
Available out of the box with no license required.
ua_precheck: Pre-flight architectural risk check (10 checks/day, default critical-path rules only)ua_status: Returns MCP health status.ua_scan: Forces a re-scan of the workspace.ua_graph_summary: Returns aggregated node/edge statistics.ua_architecture_report: Groups files by top-level modules.ua_dependency_report: Identifies files with the most incoming dependencies (fan-in).ua_explain: Retrieves 1-hop dependencies for a specific file.ua_onboarding_doc: Generates onboarding context.
Premium Tools (Pro Tier)
ua_precheck: Unlimited pre-flight checks with configurable critical paths and .ua-rules.json enforcementua_rules: Enforces custom.ua-rules.jsonboundaries.ua_rules_check: Validates the rules schema.ua_find_callers: Retrieves reverse dependencies up to 2 hops.ua_impact_analysis: Retrieves full transitive closure of reverse dependencies.ua_validate_graph: Checks the knowledge graph schema for corruption.ua_ci_check: Analyzes Git PR diffs for architectural impact.
[!IMPORTANT] The True Enforcement Backstop: While local agents rely on system instructions to run
ua_precheck, theua_ci_checktool is designed to be your unbypassable safety net. By runningua_ci_checkin your GitHub Actions and requiring it as a status check in GitHub Branch Protection, you ensure that no rogue edits can ever be merged into production without explicit approval.
Pricing
| Tier | Price | Features |
|---|---|---|
| Free | $0 forever | Basic graph operations, local storage. |
| Pro | $10/month OR $50 one-time | Unlimited nodes, advanced graph analytics, rule enforcement, priority support. (Lifetime access limited availability) |
Get your license key:
Troubleshooting
- Server fails to start: Ensure you have Node.js v18 or later installed.
- License key error: Verify your key in the
.envfile or Claude config matches the one on your dashboard. - Path not found: Ensure
UA_PROJECT_PATHis absolute or resolves correctly relative to where the server runs.
License
MIT License
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。