MCP servers that ship as code — a repository or a package you install and run locally (usually over stdio). They have no hosted endpoint, so we cannot probe or grade them: instead each one shows its repository, install command, license, stars and when it was last updated.
Collected from GitHub repositories and npm, PyPI, Cargo, Go, Maven, NuGet and RubyGems packages tagged as MCP servers, the official MCP Registry, Docker's MCP catalog, ModelScope, Baidu MCP World, Cline and other public directories; one entry per repository. A server that also offers a hosted endpoint is listed under Remote instead. These entries are not counted in our reachability, verification or readiness statistics.
-
—
MIT -
gguf-mcpgithub.com/arose26/gguf-mcpupdated this year
MCP server that inspects local GGUF and safetensors model files — architecture, quantization, parameters, tensors, and VRAM-fit estimates from headers only, without loading the model
npm MIT -
本地模型管家 MCP:管理 GGUF 模型与 llama-server —— 模型台账 / 显存估算 / 服务生命周期 / chat 代理(FastMCP)
MIT -
mcp-turboquantgithub.com/ShipItAndPray/mcp-turboquantupdated this year
MCP server for LLM quantization. Compress any model to GGUF/GPTQ/AWQ in one tool call. First MCP server for model compression.
MIT ★ 5 -
unchgithub.com/uchebnick/unchupdated this year
Local-first semantic code search for repository annotations via GGUF embeddings and sqlite-vec.
go Apache-2.0 ★ 15 -
? Tiny Decision - A Local Harness & Deployment Platform for System One Models. In-process GGUF/ONNX runtime, calibrated confidence, visual DAG pipelines, and MCP export.
MIT ★ 1 -
Generate and edit images from Claude Code on your own GPU. MCP server + CLI for local GGUF models (Z-Image, FLUX.2 klein) via stable-diffusion.cpp.
Apache-2.0 -
Static security scanner for machine-learning model files. Checks PyTorch, pickle, joblib, GGUF, SafeTensors, ONNX and Keras checkpoints for code execution before you load them. CLI, GitHub Action, load guard, MCP server.
pypi MIT ★ 2 -
Multi-backend GGUF vision-language inference engine (moondream3.1 · VideoChat3 · Qwen2.5-VL · MOSS-VL · VideoMAE/ViViT · SmolVLM) with a unified CLI, Web UI and an MCP server for agents.
NOASSERTION ★ 4 -
A local studio to tune and extend language models: versioned datasets, LoRA and QLoRA training on the allowed GPUs, merges, GGUF conversion, importance-matrix quantization, YaRN context extension and evaluation against the base model. MCP tools for assistants.
MIT -
blagithub.com/qybaihe/blaactive
🛡️ BLA (bla-agents / Blind LLM Agents) — Open-source Blue Defense MCP server & CLI protecting JS/TS, Electron ASAR, WASM, Python, Native Binaries, SQLite, GGUF/Safetensors & Android DEX against AI reverse-engineering agents (REA).
npm MIT ★ 5,810 -
dsh-unslothactive
Unsloth for DeepSeek Harness: local LLM fine-tuning and dataset generation. Bundles a zero-dependency Python MCP server and mounts DSH's own MCP client, exposing model inventory, Data Recipe dataset generation, LoRA/QLoRA training, GGUF export and inferen
npm MIT -
Impulse by Velocity: run AI models on your own computer, on Windows and Linux. Load any GGUF model and chat with web search, files, charts and voice, let it use your computer, and serve it to any app through an OpenAI-compatible API and MCP. Momentum memory up to 32M tokens. No cloud.
NOASSERTION ★ 35 -
comfyui_LLM_partyygithub.com/heshengta0/comfyui_LLM_partyyupdated this year
LLM Agent Framework in ComfyUI includes MCP sever, Omost,GPT-sovits, ChatTTS,GOT-OCR2.0, and FLUX prompt nodes,access to Feishu,discord,and adapts to all llms with similar openai / aisuite interfaces, such as o1,ollama, gemini, grok, qwen, GLM, deepseek, kimi,doubao. Adapted to local llms, vlm, gguf such as llama-3.3 Janus-Pro, Linkage graphRAG
AGPL-3.0 ★ 2 -
comfyui_LLM_partygithub.com/heshengta0/comfyui_LLM_partyupdated this year
LLM Agent Framework in ComfyUI includes MCP sever, Omost,GPT-sovits, ChatTTS,GOT-OCR2.0, and FLUX prompt nodes,access to Feishu,discord,and adapts to all llms with similar openai / aisuite interfaces, such as o1,ollama, gemini, grok, qwen, GLM, deepseek, kimi,doubao. Adapted to local llms, vlm, gguf such as llama-3.3 Janus-Pro, Linkage graphRAG
AGPL-3.0 ★ 2 -
comfyui_LLM_partygithub.com/heshengtao/comfyui_LLM_partyupdated this year
LLM Agent Framework in ComfyUI includes MCP sever, Omost,GPT-sovits, ChatTTS,GOT-OCR2.0, and FLUX prompt nodes,access to Feishu,discord,and adapts to all llms with similar openai / aisuite interfaces, such as o1,ollama, gemini, grok, qwen, GLM, deepseek, kimi,doubao. Adapted to local llms, vlm, gguf such as llama-3.3 Janus-Pro, Linkage graphRAG
AGPL-3.0 ★ 2,390 -
devonbike.overfit.mcpgithub.com/devonbike/overfitupdated this year
A dependency-free MCP (Model Context Protocol) stdio server for the in-process Overfit LLM runtime — typed JSON-RPC 2.0 contracts with source-generated System.Text.Json serialization, no SDK dependency, no reflection, Native-AOT verified. Exposes local, zero-egress tools to MCP hosts (Claude Code, Claude Desktop, IDEs): ask a local GGUF model, query private documents with RAG citations, transcribe audio with Whisper — all in pure .NET on the CPU. Dual-licensed (AGPL-3.0-or-later / commercial).
nuget other -
This is a local LLM provider, include rag-mcp and agent features, can load HF/gguf/onnx model formats..
Apache-2.0 -
MCP stdio server for Windsurf that routes tool calls to a local llama.cpp llama-server (GGUF), optimized for low-VRAM GPUs.
★ 1 -
vMLX - Use MLX models easily - JANGQ (GGUF for MLX) - Not dependant on mlx_vlm
pypi Apache-2.0 ★ 918 -
Turing Translate: a docked Windows desktop translator with local GGUF models or online providers, phrase memory and an MCP server
MIT -
MCP server + toolkit for the geometry of meaning — semantic drift, trajectory dynamics, and axis-alignment analysis over text embeddings (NV-Embed-v2 / GGUF backends).
★ 1 -
SharpInfergithub.com/JPaulDuncan/SharpInferupdated this year
A pure C# LLM inference engine built from scratch — no Python, no llama.cpp bindings, no ONNX Runtime. SharpInfer loads GGUF and Safetensors models directly, dequantizes weights in managed code, and runs the full transformer forward pass natively on .NET 8.
MIT ★ 6 -
ai-compute-radaractive
MCP server bridge for AI Compute Radar: which open models fit your GPU or Mac (measured GGUF sizes + computed context cache), GPU rental prices from eight sources, model momentum, weekly pick. Read-only, no auth, no telemetry.
npm MIT -
Agent-drivable generative creation studio on the desktop: 310 MCP actions across 26 families cover everything the interface can do. Local models on the machine — GGUF LLMs, offline dictation — and cloud generation of images, video and 3D, with API keys encrypted in the main process, never exposed to the renderer.
NOASSERTION ★ 1 -
compress-tokensgithub.com/Amir-Zecharia/compress-tokensupdated this year
MCP server that compresses text by removing unnecessary tokens using local LLM surprisal scoring
MIT ★ 1 -
On-device clinical documentation: speech to reviewed clinical note, fully offline. React Native, quantized MedASR and MedGemma, FastAPI, FHIR R4, MCP.
Apache-2.0 -
Profile llama.cpp-family GPU inference, run controlled experiments, and compare results with evidence labels. CLI, MCP server and Agent Skill.
MIT ★ 1 -
compressx-mcpupdated this year
MCP server for CompressX — lets Claude Code, Cursor, and other AI tools compress LLM models via the Model Context Protocol.
npm UNLICENSED -
Benchmark local LLM inference speed (tokens/sec) on your own hardware — llama.cpp native + cloud APIs, 124-model catalog, optimal-quant picker, and an MCP serve mode.
MIT ★ 2 -
Odinfer — a high-efficiency local inference engine for Qwen3.5-class models on Apple Silicon (MacBook Air first). Dense + MoE (35B-A3B family), MCP/HTTP server, zero-compile binaries, pure Odin.
MIT ★ 5 -
llama.cpp's llama-server + Web UI, compiled natively for your exact CPU/GPU in Docker, with a hardware guard and private local web search (SearXNG + MCP).
MIT ★ 2 -
The home port for your local AI. Zero-install Mac app + MCP server: talk to Claude to discover, download and verify Hugging Face models straight to your own drive. By CAIO Group.
MIT ★ 2 -
mollagithub.com/tamnd/mollaactive
Local inference server in pure Mojo. Speaks OpenAI, Anthropic, and MCP. Models are OCI artifacts. Same source runs on CPU, NVIDIA, AMD, and Apple GPUs. Apache-2.0 all the way down.
Apache-2.0 -
JevStyle: local decision models for AI agents. Classify, route and score with a probability per option. systemone-compatible local server, MCP tools, Claude Code guard. Open weights, Apache-2.0.
pypi Apache-2.0 ★ 135 -
diffusion-llm-mcpgithub.com/sandraschi/diffusion-llm-mcpupdated this year
FastMCP fleet MCP server for diffusion LMs (dLLM). DiffusionGemma on Goliath RTX 4090 — batch inference, HLE-shaped reasoning, ~200–400 tok/s. Doc phase; llama-diffusion-cli sidecar next. Complements local-llm-mcp.
MIT ★ 2 -
A locally-deployed, offline security intelligence server powered by a quantized LLM + RAG pipeline over CVE feeds and logs, exposed as an MCP server. Plugs into Claude Desktop, Cursor, or any MCP client - no cloud, no data leaks.
MIT -
Run AI models like Llama, Qwen, and Gemma on every computer you own, and manage them for everyone on your network. Local first, with OpenAI- and Anthropic-compatible APIs, people and roles, and MCP.
AGPL-3.0 ★ 5 -
Profile-based launcher for local LLM servers - start, stop and switch models on llama.cpp, Ollama and LM Studio from one CLI/TUI. Zero resident overhead, optional MCP control plane, importable as a Go library.
MIT ★ 2 -
A compiler for the code AI agents write: a flat SSA language with JSON diagnostics, compiled in under a millisecond to sandboxed native code, as MCP tools. Its Llama 3.2 1B prefills 1.7-2.4x faster than llama.cpp.
Apache-2.0 ★ 1 -
A confined generation tool: runs an LLM over a local directory read-only and emits the output as a product that is always labeled untrusted. CLI + stdio MCP server on the myelin substrate, optional in-process creatine engine. Part of the Nootropics stack.
GPL-2.0 -
light-OCR-modulegithub.com/dwain-barnes/light-OCR-moduleupdated this year
Local OCR that turns scanned PDFs and images into clean Markdown — powered by LightOnOCR + llama.cpp. 100% on-device, no cloud, no API keys. Python library, CLI, MCP server, and Claude Code skill in one package
MIT -
A local model engine with a search engine and web crawler built in. Run your coding agents on your own GPUs, and get cited answers from your files, code, and the sites you crawl. One executable.
go MIT ★ 71 -
Bulk AI image generation on your own Windows PC: FLUX.2, Qwen-Image, SDXL and more via stable-diffusion.cpp. Batches, reference images, LoRAs, upscaling, an OpenAI-compatible API and an MCP server for AI agents. No cloud, no account.
MIT -
Local AI desktop app: a Chromium browser with a GPU-aware llama.cpp runtime, OpenAI-compatible API, AI chat and agents, Stable Diffusion image/video, TTS and music, plus an MCP server so Claude Code, Codex and Cline can drive your browser. Windows, macOS, Linux.
NOASSERTION ★ 151 -
Peridot v1.6.0 [STABLE] | Sovereign AI Kernel. Offline, GPU accelerated Qwen framework: a 27B model fully on an 8GB GPU at 23.6 tok/s. Sandboxed skills & plugins, model tool use, opt in web search, MCP bridge for private data jobs from cloud agents, ZAT-SCS preemption, RAG. Zero data exfiltration. Engineered by uncoalesced.
MIT ★ 8