📡 MCP Radar

naderloocodelab/micro-actinf

⚡ Ultra-fast O(1) Active Inference & Bayesian Cognitive Governor for AI Agents (Antigravity, Claude, Cursor) & Embedded Systems. Zero-alloc C11 core (<36KB) + FastMCP server.

⭐ 5 stars
⑂ 0 forks
C
momentum ▲ 10.0
created 2026-09-13
on radar since 2026-09-17
star trend 4 → 5 since 2026-09-17
active-inferenceai-agentsantigravitybayesian-inferencec11claude-desktopcursor-aidirichlet-distributionembedded-aifree-energy-principlegame-aimcp-servermodel-context-protocolpomdproboticszero-allocation
View on GitHub ↗Homepage ↗

About micro-actinf

🧠 Micro-ActInf: Ultra-Fast $O(1)$ Active Inference & Bayesian Cognitive Governor

Eliminate context drift, hallucination loops, and conversational token bloat in AI agents (Antigravity, Claude Desktop, Cursor) with Karl Friston's Free Energy Principle and an ultra-lightweight 20KB C11 POMDP engine.

📑 Table of Contents - Why Micro-ActInf? (The Problem & Solution) - Feature Comparison vs Existing Frameworks - Dual-Target Architecture - How It Works: The Active Inference Loop - The 6 Cognitive Regimes & 4 Action Policies - Quickstart: Plug into AI Agents in 60 Seconds - 1. Google Antigravity - 2. Anthropic Claude Desktop - 3. Cursor / VS Code - 4. Pure Offline Local LLMs (Ollama / vLLM / llama.cpp) - Embedded Systems & Game Engines (C11 Core) - Mathematical Rigor - Empirical Benchmarks - 🇮🇷 راهنمای جامع فارسی (Persian Technical Guide)

💡 Why Micro-ActInf? (The Problem & Solution)

🚨 The AI Agent Crisis: 1. Context Drift & Goal Degradation: In multi-turn coding sessions, LLM agents forget high-level constraints, get distracted by minor side-tracks, and drift away from the original engineering objective. 2. Infinite Debugging Loops: When an error occurs, LLMs often attempt the same failed fix repeatedly or oscillate between conflicting implementations. 3. Conversational Token Waste: Up to 40% of generated tokens are squandered on polite filler ("Certainly!", "I'd be happy to help!"), repetitive apologies, and explanatory essays when you only needed production code. 4. Soft Constraints Fail: Prompt instructions ("be concise", "focus on tests") are probabilistic recommendations t

From the project README.

Maintaining this server?

Add the radar badge to your README — it shows your project was picked up by MCP Radar and links to this page:

[![On MCP Radar](https://mcp.liqiwa.com/badge.svg)](https://mcp.liqiwa.com/s/naderloocodelab--micro-actinf.html)

Related MCP servers

AIYantra/extra

The Open-Source Astra 6 for your PC. Your AI can see, click, type, navigate, and get real work done on Windows.

⭐ 4Python
OneInterface/stormy-cookbook

Open-source cookbook for the Stormy Social Data API and MCP server (Model Context Protocol) — one REST API for the TikTok API, YouTube API, Instagram API, LinkedIn API, X (Twitter) API and Reddit API.

⭐ 72
00200200/usagetrim

Keep the signal. Cut the noise. Local CLI + MCP — fold verbose tool output for Claude Code, Codex, Cursor & Desktop.

⭐ 11Python
aminmadaniofficial/bale-mcp-server

Production-grade Model Context Protocol (MCP) server for Bale Messenger with 32 tools for LLMs (Claude, Cursor, Antigravity) to interact with user accounts.

⭐ 7Python
awais2o/devops-deploy-guard-mcp-server

Production-ready, zero-knowledge AI DevOps deployment, infrastructure provisioning, self-healing execution engine, and runbook generator for VPS servers via Model Context Protocol (MCP).

⭐ 4TypeScript
thatcreativetayo/kromamcp

Zero-database, internet-connected MCP server for live design intelligence and UI component synthesis.

⭐ 3TypeScript

📬 Get the weekly radar in your inbox

The top new MCP servers of the week, every Monday. No spam, unsubscribe anytime.