AI Integration

Give your agent runtime context, not just source code.

BrokenApp's MCP server builds a 7-layer graph of your entire application — code, frontend, database, runtime, docs, conversations, and design — then exposes it to Claude Code, Cursor, and any MCP-compatible AI tool. Your agent doesn't just read source files. It sees how your app actually works.

Application Graph

Your AI agent sees your whole app

Every scan builds a unified graph across 7 layers. Your agent can query any layer, trace connections between them, and understand your application as a system — not a pile of files.

Code

Source files, functions, imports, and dependency graph across all languages.

Frontend

Routes, components, forms, navigation flows, and UI state machines.

Database

Tables, columns, relations, indexes, and migration history.

Runtime

API endpoints, request/response schemas, auth flows, and network traffic.

Docs

README files, inline documentation, API docs, and architecture notes.

Conversations

Chat history, agent threads, past decisions, and context from previous sessions.

Design

Accessibility violations, color contrast, spacing, and UI consistency issues.

In action

Watch the agent work

Your AI agent queries the app graph, reads critical findings, generates fixes, and re-scans to verify — all through MCP tool calls.

agent → scan

"I'll scan the codebase and check for security issues."

mcp ←

Scanned 847 files. 234 code nodes, 89 endpoints,
12 security findings stored to graph.

agent → list_findings

"Found 3 high-severity issues. Let me check the details."

mcp ←

3 high findings:
  Hardcoded JWT secret (config.ts:23)
  SQL injection (users.ts:89)
  Missing CSRF (/api/payment)

agent → scan

"I'll fix the hardcoded JWT secret first, then re-scan to verify."

mcp ←

Re-scanned. 11 security findings (was 12).
Hardcoded JWT secret resolved.

MCP Tools

What the server exposes

13 tools across 4 categories. Each tool returns structured data your agent can reason about, chain together, and act on.

Scanning

scan

Run a full scan across 6 layers: code, frontend, security, accessibility, design, and database.

Analysis

Full-text search across the entire application graph. Find endpoints, components, findings, and more.

list_findings

List all security and quality findings with severity, CWE mapping, evidence, and remediation.

graph_health

Get a health summary of the application graph: coverage, staleness, and layer completeness.

impact_analysis

Trace the blast radius of a change. Shows affected routes, components, and downstream dependencies.

endpoint_trace

Trace an endpoint from frontend form to API handler to database query. Full request lifecycle.

api_coverage

Compare discovered API endpoints against defined routes. Find untested or undocumented paths.

Management

baseline

Create, apply, or diff scan baselines. Filter noise and track regressions across runs.

ingest

Ingest external data into the graph: OpenAPI specs, Postman collections, HAR files, and more.

docs_report

Generate a documentation report from the graph. Markdown, JSON, or structured output.

Portal

portal_scan

Trigger a scan on BrokenApp Cloud. Returns scan ID for polling results remotely.

portal_projects

List and manage projects on BrokenApp Cloud. View scan history and team access.

sync_to_platform

Push local scan results and graph data to BrokenApp Cloud for team sharing and dashboards.

Setup

Install in under a minute

The MCP server ships as an npm package. No Rust toolchain or compilation required. Pick your tool and paste the config.

Claude Desktop

Add to your claude_desktop_config.json

{
  "mcpServers": {
    "brokenapp": {
      "command": "npx",
      "args": ["-y", "brokenapp-mcp"]
    }
  }
}

Cursor

Add to .cursor/mcp.json in your project root

{
  "mcpServers": {
    "brokenapp": {
      "command": "npx",
      "args": ["-y", "brokenapp-mcp"]
    }
  }
}

Claude Code

One command from your terminal

$ claude mcp add brokenapp -- npx -y brokenapp-mcp

Tell your AI what to do

"Scan this project, find all critical security issues, fix them, and re-scan to verify."

Install the MCP server. Give your agent eyes on the running app.

Works with Claude Desktop, Cursor, Claude Code, and any MCP-compatible tool.

$ npx -y brokenapp-mcp

Install CLI View docs