A

ANA MAX

Advanced Neural Architecture

local QA lab / offline agents / MCP auth on

Give local AI agents eyes, hands, and verification.

ANA MAX is a Windows-first MCP runtime for QA labs, private workstations, offline LLMs, and AI coding agents that need to see the workspace, inspect the desktop, use focused tools, instrument runtime behavior when needed, speak status, and verify the result.

Watch demo 1 Watch demo 2 90s proof script Connect MCP Run checks
64loaded tools
4premium capabilities
9AI Core adapters

Runtime WorkGraph

observe -> act -> verify
MCP Clients Claude / Cursor / Windsurf main.py MCP Server Flask / JSON-RPC 2.0 / auth required Tool Registry + License Gate 64 tools / runtime premium blocking Desktop Eyes capture / UIA / OCR Developer Tools files / git / tests / shell AI Core memory / context / orchestration WorkGraph Discipline observe first / choose useful tools / verify result
MCP Clients

The coding assistant asks for real observations instead of guessing. Cursor, Windsurf, Antigravity, OpenCode, or Codex can call ANA through MCP.

Ask the agent to inspect first, then edit only after a tool result.

Not another tool list. A workflow for local agents that should not guess.

ANA MAX combines local/offline privacy, optional online models, desktop vision, Windows UI/API awareness, git state, tests, memory, voice, and safe runtime instrumentation. It is meant for QA testers, lab operators, and AI agents that must work from local facts without exposing private work.

Observe Read files, git state, terminal output, UIA, screenshots, and logs before acting.
Instrument Use Frida or Windows inspection for authorized runtime facts when the task needs it.
Act Edit code, run commands, control UI, and coordinate tools through MCP.
Verify Run smoke tests and report what passed or failed instead of hiding uncertainty.
Public demo policy 
Public video:
https://www.youtube.com/watch?v=yIWILEd6glU
https://youtu.be/2Ft9eRO7GzM

Do not store large videos in git.
Host future demos on YouTube or GitHub Releases.
Keep the repository source-only and public-safe.
local lab workflow

See what the tools prove.

The fastest way to understand ANA MAX is to watch an agent observe the workspace, use local tools, avoid repeated mistakes, and verify the result.

  1. Observe the desktop, logs, terminal, and project state.
  2. Use focused tools for QA, debug, and offline work.
  3. Run smoke checks, speak status, and report facts.

Use the 90-second proof script and the local QA lab vision to record a sharper public demo.

Current public demos: demo 1 and demo 2.

Local server

ANA MAX starts on http://127.0.0.1:8765. MCP requests go to /mcp.

Auth required

Set MCP_API_KEY or ANA_MCP_KEY, then send Authorization: Bearer ....

Public hygiene

No private memory, logs, databases, screenshots, local paths, tokens, or license files belong in the public release.

Start ANA MAX 
$env:MCP_API_KEY = "change-me"
python main.py
List tools through MCP 
curl -X POST http://127.0.0.1:8765/mcp `
  -H "Content-Type: application/json" `
  -H "Authorization: Bearer change-me" `
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list","params":{}}'

Free Vision AI

desktop_capture is free in this release and can be used for observe and verify workflows.

Premium capabilities

Premium desktop automation and deep Windows inspection are blocked at runtime without a valid Pro license.

  • live_desktop_viewer
  • desktop_control and desktop_control_tool
  • windows_insight and windows_insight_tool
  • windows_deep_sight

Activation

Use python activate_license.py --key YOUR_KEY, then restart ANA MAX.

Expected smoke

python main.py --test should report 3 PASS / 0 FAIL.

Expected tools

python main.py --list-tools should report 64 loaded tools.

Expected tests

python -m unittest discover -s tests -v should pass all tests.

Release verification 
python -m compileall -q main.py core tools vscode_extension
python main.py --test
python main.py --list-tools
python -m unittest discover -s tests -v