MCP servers ×21 CrewAI Ollama local ₹0 inference FalkorDB

Open for freelance AI work

Agents that
think for ₹0.

Anuj — building M4ST, a local-first multi-agent AI operator stack, from Bareilly, India. Zero cloud spend. Self-taught, self-hosted, shipped not planned.

See the systems → Hire me

21 MCP SERVERS11 FREE LLM PROVIDERS₹0 CLOUD COST12MS CACHE LATENCY40% COST CUT 21 MCP SERVERS11 FREE LLM PROVIDERS₹0 CLOUD COST12MS CACHE LATENCY40% COST CUT

00 — ORIGIN

No GPU cluster. Just obsession.

I started with zero resources — no funded infra, no premium APIs, no team. Just a laptop and a conviction that intelligent systems shouldn't need a data center.

Everything in M4ST is self-taught, self-funded, self-hosted. While others provisioned GPU clusters, I routed across free-tier providers to keep inference at zero cost — without sacrificing reliability.

Currently chasing CEH certification while taking on freelance AI/automation work. Jugaad-first: shipped beats planned, always.

Constraints breed creativity. Limited resources forced me to optimize everything. That's not a limitation — that's an edge.

— Anuj, on M4ST

01 — ARSENAL

What I actually build with.

Multi-Agent Orchestration

Autonomous agent swarms that collaborate and delegate — no human in the loop.

CrewAILangGraphn8nAutoGen

MCP Protocol Engineering

21 custom servers bridging LLMs with databases, browsers, OSINT tools.

Zero-Cost Routing

11 free providers, ₹0 inference.

Local Inference

Ollama + quantized GGUF on consumer hardware.

Semantic Memory

12ms cache, 40% fewer calls.

Offensive & Defensive Security

CEH-track OSINT, threat modeling, recon pipelines — applied to real infra audits.

02 — SHIPPED

Systems in production, not slideware.

M4ST Stack Live

The complete autonomous agent operator system — 5 agents, MCP orchestration, semantic caching, zero-cost LLM routing.

CrewAIMCPFalkorDB

OpenWork Integration Active

Agentic coding workspace wired into GitHub and Vercel MCP for direct build-and-deploy.

MCPGitHub

cai-osint Open Source

OSINT framework — automated data gathering, Shodan integration, AI-powered analysis.

PythonShodan

Semantic Cache Engine v2.0

Sub-12ms caching over ChromaDB embeddings, 40% fewer LLM calls.

ChromaDBPython

03 — TIMELINE

How M4ST actually happened.

2024

Zero-cost breakthrough

Mastered free-tier LLM routing across 11 providers, hitting ₹0 inference cost.

2025

MCP revolution

Built the first custom MCP server, expanded rapidly to 21 — databases, browsers, OSINT.

2026

M4STCLAW v3

First working agent stack — obsession begins, iterating through v5, v6, v7 on nights and weekends.