MCP servers and skills: search, filter by where they run, and compare.
Layered Mixture-of-Agents consensus synthesizer & quality amplifier (Together MoA style)
SkillLink only★ 8
🦀⚡ Community skill pack & CLI for Genspark Claw, OpenClaw & Hermes agents — 5 production-ready SKILL.md skills (deep research, reports, slides, browser automation, AI call prep), zero-dependency Node
Node.jsMIT★ 538No key
Sparse Mixture-of-Experts (MoE) router and code reasoning engine (Mistral style)
SkillLink only★ 8
Metis (μῆτις) — distributed cognitive layer over any LLM. Understanding Council → confidence gate (fail-closed) → layered MoA → verifier-with-retry, with DGPD depth gating, memory, economy metering, a
PythonMIT★ 3Few starsNo key
Mixture-of-Agents vote arbiter for consensus answer
SkillLink only★ 9
Mixture — a deliberately-small, layered skills + agents framework for Claude Code (skills, session memory, multi-agent task runtime). Install via npx.
SkillLink onlyFew stars
PyThermoModels thermodynamic calculations exposed through MCP.
PythonApache-2.0★ 1Few starsNo key
Fully local AI coding workstation: a 35B MoE model (Tiel-Coder) at 200k context on a single 12 GB GPU, 35–40 tok/s, without slowing your desktop. Tuned llama.cpp, MoE expert cache, TurboQuant fix, per
SkillMIT★ 4Few starsNo key
Substrate-based context compaction for Claude Code. Inspectable signal weights, local-only, MCP-ready. Alternative to /compact + hook-based capture (claude-mem, etc.).
PythonMIT★ 5No key
🧪 Experiments with an EU AI Act compliance MCP server and a consensus router for Mixture-of-Experts routing — a designed tally; the consensus-strength claim was retracted 2026-07-29.
PythonApache-2.0Few starsNo key
A color consultant for painting, illustration, and design: build palettes, troubleshoot muddy mixtures, and turn descriptions like “dusty blue” into practical guidance. Distinguish how colors interact
SkillMIT★ 2Few starsNo key
Can I run it? An MCP server and TypeScript library that check whether an open LLM fits your GPU: VRAM, KV cache, speed ceilings and what fits instead, with nodegrove.io's math.
Node.jsMITFew starsNo key
A mixture-of-experts (MoE) visual question-answering (VQA) server implemented as a model context protocol (MCP).
MCPLink only★ 2Few stars
Deployment-blocking bugs, a Looped-Transformer memory bound, and a chunked-prefill fix for running Nanbeige4.2-3B on Apple Silicon (MPS)
SkillMIT★ 1Few starsNo key
A collection of Model Context Protocol (MCP) servers for chemical engineering and chemistry applications, built on top of the powerful MoziChem framework.
PythonMIT★ 4Few starsNo key
An MCP fold tool: agents run sub-tasks in isolated branches and get back clean summaries, so the main context never fills up. Context folding, no RL.
Node.jsMIT★ 1Few starsNo key
AI agents that improve themselves. Self-evolving prompts via A/B testing, multi-model critics, and safety gates. TypeScript. Zero config.
Node.jsMIT★ 10No key
Massive multi-source academic literature search for Claude Code — one skill fans out subagents across 20+ scholarly databases (arXiv, Semantic Scholar, Crossref, OpenAlex, PubMed, …), merges into a de
SkillMIT★ 31No key
Hear a track from a URL or search phrase: acoustic profile + plain-language description, a music_v2 block (rhythm/groove, harmony, timbre, 64-dim similarity embedding), and a critic block for genre /
SkillMIT★ 4Few starsNo key
A modular Python toolkit for creating MCP modules tailored to chemical engineering and chemistry workflows.
SkillMIT★ 7No key
Free, open-source SMILES → IUPAC name converter. PubChem-keyed lookup with OPSIN round-trip cross-validation. CLI, Python lib, Gradio web app, MCP server.
PythonMIT★ 2Few starsNo key
Who stands behind it
