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.
-
vllm-mcp-servergithub.com/micytao/vllm-mcp-serverupdated this year
MCP server for vLLM - expose vLLM capabilities to AI assistants
pypi Apache-2.0 -
vllm-ci-mcpgithub.com/dougbtv/vllm-ci-mcpupdated this year
MCP Server for vLLM CI constext
Apache-2.0 ★ 1 -
A very simple proof-of-concept mcp for running vllm benchmarks
★ 4 -
Keeps a vLLM container from overheating an NVIDIA DGX Spark, and lets AI agents read their hardware temperature over MCP
pypi Apache-2.0 -
vllm-ops-mcpgithub.com/jaimenbell/vllm-ops-mcpupdated this year
Read-only ops/health MCP server for a WSL2/systemd vLLM instance: liveness vs real-completion health tiers, GPU/VRAM, systemd state, live serve-flag introspection.
pypi MIT -
Governed GPU inference ops (vLLM + Ray Serve): latency RCA, scaling, drain, 30 MCP tools (preview)
pypi MIT -
MCP server that delegates Claude Code agent work to local models (llama.cpp / vLLM) on your own hardware
MIT ★ 19 -
MCP server for a team's model & service catalog (vLLM, embeddings, custom services) with REST API and web UI
MIT -
MCP server for conversation history search and retrieval across Claude Code, Antigravity, Windsurf, Roo Code, Codex, and vLLM
npm MIT -
Serve, size and benchmark open LLMs on one 16 GB GPU: vLLM vs llama.cpp vs Ollama, KV cache and offload planning, MCP server
MIT -
Sandboxed local coding agents for Claude Code: an MCP server that runs Claude Code on your own GPU (vLLM / Ollama) in hardened, network-less containers
Apache-2.0 -
Denpex MCP Server: Diagnose GPU & ML training failures (CUDA OOM, NCCL timeouts, NVIDIA Xid, Slurm, vLLM) from Claude Desktop, Cursor, Windsurf, and Cline.
-
Transform LinkedIn saved posts into an interactive, interconnected Obsidian knowledge graph. Features a zero-config web dashboard, hybrid LLM routing (Groq + local vLLM/llama.cpp), vector embeddings, and MCP support.
MIT -
AI-агент, который выполняет задачи в браузере через function calling, и MCP-сервер с инструментами браузера. Работает с OpenAI и self-hosted моделями через vLLM
-
Real-time and historical metrics for locally served LLMs (Ollama and vLLM) — dashboard, HTTP API and MCP server. One process, one SQLite file, never in the request path.
Apache-2.0 ★ 1 -
DIOgithub.com/nisaral/DIOactive
Drop-in OpenAI- and Ollama-compatible LLM gateway that learns each backend's latency online and routes vLLM / SGLang / TGI / Ollama with SLO-aware admission. No engine patches.
Apache-2.0 ★ 3 -
grepaigithub.com/vp2722/grepaiupdated this year
🔍 GrepAI - Local AI-powered codebase search MCP Server. Self-hosted MCP server with GPU-accelerated LLM support (llama.cpp/vLLM). Search, analyze, and navigate your codebase with Claude Code or any MCP-compatible AI agent. One-command installation.
MIT -
pdf4vllm-mcpgithub.com/PyJudge/pdf4vllm-mcpupdated this year
PDF reading MCP server optimized for vLLMs
pypi NOASSERTION -
modelroutegithub.com/cognis-digital/modelrouteupdated this year
Local model router / proxy across Ollama, vLLM, and cloud with fallback
NOASSERTION -
Claude Code plugin: MCP server that offloads bounded LLM tasks to cheaper local (LM Studio, Ollama, vLLM, llama.cpp) or remote (OpenRouter) models with profile-based configuration and ensemble mode
MIT ★ 1 -
local-llm-delegategithub.com/faizan007jr/local-llm-delegateupdated this year
Claude Code plugin that delegates simple, well-scoped tasks to a locally running LLM (Ollama, LM Studio, llama.cpp, vLLM) via MCP
MIT ★ 1 -
Live2D multi-agent animation prototype with MCP, A2A, provider routing, local AI, ComfyUI/Ollama/vLLM, and FFmpeg.
Apache-2.0 ★ 2 -
Bridge any OpenAI-compatible API (DeepSeek / Zhipu / Qwen / Moonshot / vLLM) into a Model Context Protocol server for Claude Desktop, Cursor and other MCP clients.
MIT -
openai-compat-mcpactive
Bridge any OpenAI-compatible API (DeepSeek / Zhipu / Qwen / Moonshot / self-hosted vLLM) into a Model Context Protocol server for Claude Desktop, Cursor and other MCP clients.
npm MIT -
⚡ MCP Server & API for real-time B2B buying signals, tech stack adoption (Rust, vLLM, CUDA) & AI SDR pitch angles from top AI companies.
-
Self-hosted DeepSeek-OCR-2 WebUI & API — vLLM / transformers on NVIDIA, Apple Silicon & CPU, per-page quality checks, PDF streaming, MCP
MIT ★ 447 -
JigsawStack vOCRno update data
OCR + AI = Magic: Extract data from any image in a consistent structure with fine-tuned vLLMs for the highest accuracy. Learn more about AI backed OCR here: https://jigsawstack.com/docs/api-reference/ai/vocr Get your jigsawstack API key here: https://jigsawstack.com/dashboard/
smithery -
Self-hosted VLM image-recognition MCP server (stdio). Paste an image in Claude Code / Codex / OpenCode → get recognition. OpenAI-compatible backend (vLLM / Ollama / GLM-4V / Qwen-VL).
MIT ★ 1 -
MCP server that estimates, trims and compacts context to keep any OpenAI-compatible endpoint (oMLX, Ollama, llama.cpp, LM Studio, vLLM, OpenAI) under its context window.
pypi MIT -
MCP server that saves Claude Code tokens by delegating bounded tasks to local or cloud LLMs. Works with LM Studio, Ollama, vLLM, DeepSeek, Groq, Cerebras.
npm Apache-2.0 ★ 122 -
terradev-mcpgithub.com/theoddden/terradev-mcpupdated this year
Complete Agentic GPU Infrastructure for Claude Code — 192 MCP tools: Full training lifecycle, inference deployment with cost guardrails, Ray cluster management (Wide-EP, disaggregated P/D, parallelism strategy), vLLM lifecycle (start/stop/sleep/wake/infer
npm MIT -
Open-source, self-hosted AI platform for teams. Run private AI chat, manage Ray/vLLM GPU clusters, and connect apps and agents through an OpenAI-compatible LLM gateway and MCP.
Apache-2.0 ★ 2 -
A production-grade Model Context Protocol (MCP) server that augments local LLM inference (via vLLM or Ollama) with real-time external data (e.g., weather, finance, news) to make responses dynamic, accurate, and context-aware. This system is designed for modularity, scalability, and performance in local or hybrid LLM deployments with support for co
-
AI browser agent in Chrome's side panel, powered by a model on your own PC (Ollama, LM Studio, llama.cpp, vLLM). No account, no telemetry. Opt-in CAPTCHA solver, scheduled tasks, /test bug reports, and an MCP server for Claude Code.
MIT -
ingest-apigithub.com/Perry-Multi-LLM/ingest-apiupdated this year
FastAPI-based multi-agent orchestrator for Odoo. Uses a Semantic Router to dispatch intents to specialized agents via Model Context Protocol (MCP/FastMCP). Features Guardrails AI for hallucination prevention, Human-in-the-Loop validation, Logfire observability and support for Groq, Ollama and vLLM.
★ 1 -
croit AIplane — one plane for all your AI. Self-hosted AI infrastructure connecting applications and users with models, agents, tools and enterprise data: an OpenAI-compatible LLM gateway, server-side agent execution, MCP, RAG, memory, sandbox, identity, RBAC and quotas. Routes across vLLM, SGLang, Ollama and hosted APIs.
AGPL-3.0 ★ 34 -
Two Model Context Protocol (MCP) servers for self-hosted, local AI: manage and train Ollama/vLLM models, and route questions and tasks to local LLMs through Claude, ChatGPT or local models. Use the compute of your existing hardware, and let the models tap data on your machines when you wire them in.
pypi MIT -
local-llm-mcpgithub.com/HenryLinyy/local-llm-mcpupdated this year
MCP server that lets Claude Code & Codex delegate bounded, token-heavy generation to local or cheap-cloud LLMs — LM Studio · Ollama · vLLM · llama.cpp · DS4, plus DeepSeek/OpenRouter/Groq/Cerebras/MiniMax/Agnes. Keep the frontier agent's judgment, cut the token bill, with RAM guards and a text-only safety boundary.
MIT ★ 3 -
Badgr — GPU Compute, AI Model Deployment & Agent Executionno update data
**Run GPU workloads, deploy AI models, access inference APIs, and execute AI agents using Badgr's MCP server.** Badgr is an AI compute and workload execution platform for developers, coding agents, and AI assistants. Run workloads on cloud GPUs or compatible customer-owned GPUs (BYO GPU), with cost controls, execution logs, and verified results. **What can Badgr do?** - **GPU Compute / AI Workloads:** Run Python scripts, machine learning jobs, batch inference, and supported ComfyUI workflows. - **AI Model Deployment:** Deploy and serve LLMs using compatible runtimes such as vLLM and SGLang. Get a verified, OpenAI-compatible inference endpoint. - **AI API / LLM Inference:** Make AI API requests with response validation, supported fallback routing, and bounded execution. - **AI Agent Execution:** Launch coding and testing agents, run tasks, check acceptance criteria, and retrieve execution receipts. **Key features:** Cloud GPU execution, BYO GPU support, model serving, AI agent jobs, workload verification, execution logs, cost limits, and deployment management. Connect Badgr to your MCP-compatible AI assistant to execute real workloads instead of just generating instructions. Get started: https://aibadgr.com **Requires a Badgr API key. GPU workloads and other paid execution may incur usage charges.**
smithery -
AiDeploymentMcpgithub.com/milutin2002/AiDeploymentMcpupdated this year
—
-
cwm-mcpgithub.com/mcp-tool-shop-org/context-window-managerupdated this year
MCP server for lossless LLM context restoration via KV cache persistence
pypi MIT -
mcp-kv-contextgithub.com/mcp-tool-shop/context-window-managerupdated this year
MCP server for lossless LLM context restoration via KV cache persistence
pypi MIT -
Self-hosted Qwen3 retrieval server with Embedding, Reranker, MCP, and Web console.
★ 7 -
ComputeDeck:多节点 GPU 控制面(Web / CLI / MCP)。命令行 gpu-control · AGPL-3.0
AGPL-3.0 ★ 2 -
Agent-native bi-temporal graph store: verified temporal operators as LLM tools + trace-grounded answer verification
pypi Apache-2.0 ★ 5 -
Multi-agent CTI platform — LangGraph + MCP + RAG + STIX KG + human-in-the-loop feedback
★ 1 -
atelyagithub.com/ronaldsit/atelyaupdated this year
AI-first memory OS: semantic recall that returns reusable KV. One lifecycle memory, three front doors (self-host serve, closed-LLM proxy, MCP server).
pypi Apache-2.0 -
Inspectable AI agent memory, context engineering and multi-agent orchestration with an MCP server and logical-fleet benchmarks
Apache-2.0 ★ 1 -
manimagithub.com/groscy/manimaupdated this year
Local-first MCP server that renders (and optionally generates, via Apertus 8B) Manim Community Edition animations — sandboxed, spec-driven, render-only by default.
MIT -
Give text-only LLMs eyes: a proxy swaps images/PDFs for stable text references, and an MCP ask_image tool lets the model question a separate vision model.
MIT -
Local LLM does Claude's bulk work: reads logs and files, researches the web, writes test-gated code.
npm MIT ★ 2 -
Local LLM does Claude's bulk work: reads logs and files, researches the web, writes test-gated code.
npm MIT -
Permissions-aware RAG reference implementation over Canadian federal public documents. Docker Model Runner, pgvector, Compose, Mattermost and MCP front doors, inference benchmarked locally and on Baseten.
MIT -
MCP server that hands private-data or large-output work to a local LLM: the cloud model gets sanitized digests and stable placeholders, the material never leaves your network
MIT ★ 1 -
A factory floor for AI agents: a tiny, single-binary, multi-tenant runtime that runs thousands of users' LLM agents in one process. MCP, OpenAI API, webhooks, cron, email.
Apache-2.0 -
@stealthbob80/mcp-llmupdated this year
MCP server for local LLMs — connects to LM Studio or any OpenAI-compatible endpoint. Fork of @houtini/lm with extended heartbeat and timeout handling.
npm Apache-2.0 -
mcp-anseractive
Universal 245K local Qwen3.8-27B Anser agent microkernel & MCP server: collaborative Socratic pair-programming, native multi-provider research (Brave, Tavily, Context7 docs, SearXNG, DuckDuckGo), cooperative landing turn ceiling preservation, in-process s
npm AGPL-3.0-or-later -
llm-localfirstgithub.com/shaxzodbek-uzb/llm-localfirstupdated this year
Local-first LLM router — keep sensitive data and bulk text labor on your own models; call the cloud only for the hard part. Fail-closed privacy routing + manager-worker delegation, MCP-native.
pypi MIT -
A stealth, JavaScript-rendering web browser and research sub-agents for AI agents over MCP, in one Docker container. Full logging, a live dashboard, LM Studio integration.
Apache-2.0 ★ 3 -
MCP server for measured local-LLM benchmarks from the AGmind claim registry — AMD Strix Halo (Ryzen AI Max+ 395) llama.cpp numbers today, NVIDIA DGX Spark on the bench. search_claims, get_claim, list_measured. Read-only, CC BY 4.0 data.
Apache-2.0 ★ 4