ULTRON

MCP_SERVER: ONLINEVERSION: 8bLOGIN

LOCAL-FIRST MULTI-AGENT STATIC SECURITY SYSTEM

ULTRON
Security Analysis Engine

+-------------------------------------------------------------------------+ | [AST] => [SSA_IR] => [CALL_GRAPH] => [TAINT_FLOW] => [RULES] => [LLM_AGENT] | +-------------------------------------------------------------------------+

A local-first multi-agent system that finds security flaws in source code — combining tree-sitter AST analysis, IR-based taint propagation, and specialized LLM agents.

USE IN BROWSER

LIVE_FEED

ULTRON_DAEMON_SYS_MONITOR

[!] Initializing analyzer pipeline...

[i] Mapping package boundaries: Python 3.10+

[i] Nodes generated: 8,421 declarations parsed

[i] Generating control flow graphs... DONE

[i] Running 24 taint queries on database sources

[CRITICAL] Taint path discovered in authController.ts

- SOURCE: req.body.token (line 22)

- SINK: process.execSync() (line 54)

[+] Spawning LLM Agent for verification checks...

- Agent verdict: VULNERABILITY CONFIRMED (98% confidence)

END-TO-END EXECUTION FLOW

AI Model Pipeline Architecture

Visualizing the 8 execution stages from initial repository clone to autonomous LLM verification and report synthesis.

CloneRepository Git Checkout & Workspace Init

STAGE 01

Convert to ASTTree-Sitter Multi-Language Grammar Parsing

STAGE 02

IR Pipeline3AC Single Static Assignment Translation

STAGE 03

Dependency GraphCross-File Symbol Resolution & Call Links

STAGE 04

Taint PathBackward Propagation from Sinks to Sources

STAGE 05

Static AnalysisDeterministic Rule Signatures & AST Audits

STAGE 06

Dynamic AnalysisAutonomous LLM Verification & Agentic Loop

STAGE 07

ReportExecutive PDF, Markdown & JSON Dataset Export

STAGE 08

Advanced Security Engineering Capabilities

Ultron is built with premium compilation-level analyzers that outperform standard pattern scanners.

AST Parsing

Parses codebases into concrete AST formats mapping symbols across packages.

IR Pipeline

Converts source structures to Single Static Assignment (SSA) Intermediate Representation.

Call Graph

Maps cross-module dependencies and intra-function invocation hierarchies.

Taint Analysis

Traces untrusted user payloads backwards from unsafe database or shell sinks.

Security Graph

Constructs topological model of API routers, middlewares, and backend persistence layer.

Rule Engine

Executes semantic queries and AST signature patterns to trap known vulnerability classes.

LLM Verification

Leverages localized AI agents to filter scanner noise and confirm zero false-positives.

REST API

Interactive API sandbox designed for SOC automation scripting and triggers.

MCP Server

Seamlessly exposes codebase scanning tools directly to Claude Desktop and Cursor.

1. LOCAL ENGINE QUICKSTART & CLI SETUP

Run Ultron natively on Python 3.10+ with local Ollama or cloud models.

python-3.10+tests 86 passinglicense MITlocal-first ✓

Quickstart Terminal Execution:

# 1. Install dependencies (tree-sitter, httpx, graphviz)

pip install -r requirements.txt

# 2. Run security analysis on a Git repository

python ultron.py https://github.com/user/repo

# 3. Pass --fix to invoke specialized LLM Refactoring Agents

python ultron.py scan . --fix

Ultron CLI Command Suite

Command	Description
ultron <url>	Clone a repository and run full security analysis
ultron scan <name-or-path> [--fix]	Run analysis on existing repo or local folder (pass --fix to invoke LLM Auto-Fixer)
ultron install-hook [dir]	Install built-in Git pre-commit security hook into .git/hooks/pre-commit
ultron list	List all cloned repositories
ultron delete <name>	Delete a cloned repository and workspace
ultron visualise <name>	Build & open dependency/taint/security SVGs
ultron config	Show current engine configuration

Pre-Commit Git Security Hook Integration

Install a built-in pre-commit security filter directly inside your project folder:

python ultron.py install-hook .

Bypass one-time commits using git commit --no-verify or ULTRON_DISABLE_HOOK=1.

2. MODEL CONTEXT PROTOCOL (MCP) SERVER

Expose 18 security tools directly to Claude Desktop, Cursor, opencode, and VS Code Copilot Agent.

Option A: Stdio Transport (Desktop Clients)

# Launch stdio MCP server daemon

python routes/mcp_server.py

Compatible with opencode, Claude Desktop, Cursor, and Antigravity

Option B: SSE Transport (Web Clients)

# Launch SSE MCP server on port 8743

python routes/mcp_server.py --sse --port 8743

SSE Endpoint: http://localhost:8743/sse

Claude Desktop & Cursor Config (`claude_desktop_config.json`):

{
  "mcpServers": {
    "ultron": {
      "command": "python",
      "args": ["routes/mcp_server.py"]
    }
  }
}

Exposed FastMCP Security Tools (18 Active Tools)

Tool Name	Category	Description
ultron_list_repos	Repository	List all cloned repositories with analysis status
ultron_clone_repo	Repository	Clone a Git repo and run full security analysis
ultron_scan_repo	Repository	Re-run full analysis on an existing clone
ultron_run_ast_parse	Analysis	Parse all source files into an AST using tree-sitter
ultron_run_rules	Analysis	Run deterministic rules (SQLi, path traversal, SSRF, etc.)
ultron_run_llm_detection	Analysis	Run LLM-powered vulnerability detection agent
ultron_get_findings	Results	Get cached security findings from a previous scan
ultron_get_security_graph	Results	Get full cached security graph (flows, subgraphs, summary)
ultron_get_config	Configuration	Show full configuration (ultron_config.json)

ULTRON Security Analysis Engine