🛡️ MCP Local Auditor & System Architect V2

A dual-mode, zero-token-cost local engineering engine built on the Model Context Protocol (MCP). It dynamically switches between Production Code Auditor (enforcing 10 strict engineering pillars) and System Architect & Strategist (generating topologies and execution plans) via local models in LM Studio, backed by a fast AST security filter, dual data sinks (SQLite & DPO-ready JSON logs), and a real-time Discord notification daemon.

🧩 The MCP Dual-Role Pipeline	⚙️ Engineering Autonomous Reliability
A framework for synchronized, context-aware AI workflows	Pairs an Executor with a Critic in a localized environment

🌉 The Standardized Context Bridge	⚖️ Separation of Concerns: Dual-Role Dynamic
Decoupled Read/Write operations via universal MCP layer	Executor (generation) vs Reviewer (zero-temperature validation)

       [ User Submission / Agent Request via MCP ]
                            │
                            ▼
              [ Stage 1: Fast AST Sentinel ] ─────────────────────────┐
         (Syntax Parse & Danger Call Interception)                    │
                            │ (Valid Code or Specs)                   │ (Syntax / Security
                            ▼                                         │  Violation)
           [ Stage 2: Hybrid Dual-Role LLM ]                          │
           (LM Studio Local Inference Server)                         │
            ├── Role 1: 10-Pillar Code Audit                          │
            └── Role 2: Architecture & Master Plan                    │
                            │                                         │
                            ▼                                         │
               [ Stage 3: Multi-Sink Logger ] <───────────────────────┘
            ├── SQLite DB (Metrics & Analytics)
            └── Atomic JSON Logs (DPO Datasets)
                            │
                            ▼ (File Write Event)
             [ Daemon: Discord Watcher Loop ]
         (Rich Embed Notifications pushed to Webhook)

The engine routes tasks automatically based on the payload structure passed to audit_submission:

Audits incoming implementations against 10 strict defensive engineering rules:

• Logic & Edge Conditions: Null checks, division by zero, empty collections.
• Resource Leaks & Performance: Unclosed context managers, algorithmic bottlenecks.
• Defensive Boundaries: Type hints, data contracts via dataclasses, bounded loop guards.
• Deterministic Contract: Enforces a rigid feedback schema (STATUS: APPROVED or STATUS: REJECTED).

🗺️ Pipeline Architecture Mapping	⏱️ Chronological Execution Workflow
Orchestrator routing across MCP server and dual agents	Ingest → Query → Synthesize → Audit → Output pipeline

Generates system designs for technical prompts without code:

• Component Topologies: Data flow patterns, directory structures, modular interconnects.
• Preemptive Risk Modeling: Concurrency bottlenecks, race conditions, failure points.
• Master Execution Plans: Step-by-step phased roadmaps ready for direct implementation.

watch_audit_logs.py runs as an asynchronous observer over the audit_logs/ directory, dispatching notifications whenever new audit artifacts are produced:

python watch_audit_logs.py

• Automated Polling: Tracks .json additions with debounce guards.
• Markdown Formatting: Renders status tags (APPROVED, REJECTED), duration metrics, summaries, and issue breakdowns.
• Zero Configuration Fallback: Operates headlessly in the background without halting the core MCP pipeline.

🔄 The Autonomous Refinement Engine	📦 Data Routing & State Management Payload
Continuous feedback loop eliminating hallucination drift	Strict turn-based JSON schema preventing infinite loops

• Python 3.10+
• LM Studio running an OpenAI-compatible Local Server at http://localhost:1234/v1
• Any MCP-compatible host client

# Clone the repository
git clone [https://github.com/xTanThaix/mcp-local-auditor.git](https://github.com/xTanThaix/mcp-local-auditor.git)
cd mcp-local-auditor

# Set up virtual environment
python -m venv .venv
source .venv/bin/activate  # On Windows: .venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

Add the server definition into your client's mcpServers configuration (e.g., mcp_config.json):

{
  "mcpServers": {
    "lmstudio-auditor": {
      "command": "python",
      "args": ["/path/to/mcp-local-auditor/auditor_bridge.py"]
    }
  }
}

1. Launch LM Studio and load an instruction-tuned model (e.g., Qwen-2.5-Coder, DeepSeek-Coder).
2. Open the Local Server tab.
3. Set the port (default: 1234) and click Start Server.

🏗️ Deployment Dependencies & Infrastructure	🚀 Pipeline Characteristics & Extensibility
Modular runtime stack connecting local MCP servers with LLMs	Guaranteed context alignment and drop-in extensibility

To make your coding assistant iteratively fix its own mistakes, inject this instruction rule into your agent configuration:

# Mandatory MCP Code Audit Protocol
Every time code is generated or refactored, you MUST invoke the `audit_submission` tool:
- Arguments: `task_goal`, `output_content`, `strict_rules`
- If STATUS == "REJECTED": Inspect `ACTIONABLE_FEEDBACK`, remediate reported issues, and re-submit.
- If STATUS == "APPROVED": Deliver final verified code to the user.
- Cap recovery loops at a maximum of 5 attempts.

Run the full pytest suite to validate AST guards, SQLite migrations, and mocked inference pipelines without needing a live LM Studio instance:

pytest test_auditor.py -v

If this project helps streamline your local AI engineering workflow, consider supporting development:

• Ko-fi: https://ko-fi.com/xtanthaix
• GitHub Sponsors: Available directly via repository profile

Distributed under the MIT License. Free for personal, commercial, and open-source implementation.