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.
-
Convert ONNX models to Apple Core AI (.aimodel) for on-device inference on Apple silicon. 143 op lowerings, precision checks, Python API, CLI, and MCP.
BSD-3-Clause ★ 5 -
piper-mcp-servergithub.com/nizovtsevnv/piper-mcp-serverupdated this year
Text-to-speech MCP server using Piper ONNX models
cargo MIT -
MCP server for multilingual place/location extraction (cities, countries, regions, landmarks) in 13 languages, backed by mDeBERTa/ONNX models.
MIT -
MCP server and MLOps CLI to export, quantize, audit, benchmark and auto-tune PyTorch and ONNX computer-vision models for edge deployment.
MIT -
ONNX Runtime stem separation (Demucs HTDemucs). Separate vocals, drums, bass, other from any audio file. Model auto-downloads from HuggingFace. Companion to mixx-dj-mcp.
MIT ★ 1 -
Agent-Speakergithub.com/NetDevAutomate/Agent-Speakerupdated this year
A small mcp/tool for agents (Kiro CLI, Claude Code and others...) to speak responses using a wrapper for kokoro-onnx (https://github.com/thewh1teagle/kokoro-onnx) — an 82M parameter ONNX TTS model
★ 6 -
Self-hosted real-time X stream intelligence: DAG pipelines of local ONNX models + Ollama LLMs, live map, entity graph, alerts, MCP server. Go + Vue.
MIT -
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 -
? 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 -
Local, bilingual (ES/EN) semantic search over a technical book library for coding assistants: SQLite FTS5 + ONNX embeddings, a fail-open prompt hook and an MCP server. Includes a chunking and embedding-model study (docs/paper.md).
Apache-2.0 -
This is a local LLM provider, include rag-mcp and agent features, can load HF/gguf/onnx model formats..
Apache-2.0 -
The Pure Rust Model Context Protocol (MCP) server for AI coding agents. Grounding swarms in deterministic AST graphs, BM25, and local ONNX vectors.
MIT -
AegisGate MCP — Standalone Model Context Protocol security server with 23 security layers, vendored ONNX runtime, neural threat detection (L3). SSE streaming. Session management. Zero module dependencies. Multi-arch (amd64/arm64). Apache 2.0.
oci Apache-2.0 -
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 -
MCP server that indexes a reference C# codebase with Roslyn and answers AI agents with its own idioms: written rules checked against the code, search by meaning over a local ONNX model and sqlite-vec, DI and DTO conventions. Stdio or HTTP, .NET 10.
MIT -
FaceTron is a high-performance face embedding server using ONNX Runtime, supporting dynamic multi-model loading, offline deployment, and scalable environments. It exposes an OpenAPI endpoint with MCP-compatible metadata and integrates with OpenTelemetry for observability.
MIT ★ 5 -
Local RAG server for Claude Code, Cursor and other MCP clients: hybrid BM25 + vector search with cross-encoder reranking on by default, 13 MCP tools, 33 file types including code, re-indexing on file changes. One Python process with embedded ChromaDB and FastEmbed ONNX: no Docker, database server or API key; offline after the first model download.
npm MIT ★ 306 -
Local-first MCP server for persistent semantic memory, backed by an HNSW vector index and an on-device ONNX embedder
MIT -
Personal, local-first voice and video studio for macOS: dictation, text to speech, auto-clipper, timeline editor, memory and MCP. Rust + GPUI.
Apache-2.0 ★ 1 -
Local-first embedding server (ONNX on-device or API) — the reference /embed server for CPersona. MIT.
pypi MIT ★ 1 -
Long-term memory for AI assistants, local. Rust + Qdrant + ONNX, served over MCP. Hybrid search with reranking.
npm Apache-2.0 ★ 10 -
MCP server in Rust for AI agent persistent memory: branch-aware session handoffs, local ONNX embeddings, SQLite-backed semantic search.
cargo Apache-2.0 ★ 19 -
Rust MCP server for generative media with avatars: InsightFace/ArcFace identity core, ONNX face swap, budget-guarded pipeline. Work in progress.
NOASSERTION -
DeepFetchgithub.com/vinay9986/DeepFetchupdated this year
Semantic web search MCP server with Kagi, Scrapfly, PDF extraction, and local ONNX reranking for Claude Desktop, Gemini CLI, and Codex.
Apache-2.0 -
MCP server for the SensorForge Optimizer: shrink trained models to fit edge devices, from your coding agent.
Apache-2.0 -
braingithub.com/jacquesmyo/brainupdated this year
Brain MCP server: vault search, embeddings, ingestion, calendar, whisper — second-brain backend
MIT ★ 1 -
Drive the Pollen Robotics Microduck (sim) from any MCP client — with a live Agent Experience debug page
Apache-2.0 ★ 3 -
Self-hosted MCP memory server with semantic search, swappable embedding models, and built-in benchmarking
MIT -
Persistent, local-first memory for AI agents. Selective by design: filters what's worth keeping, links what's related, and says nothing when it doesn't know. SQLite + ONNX, no cloud, no API keys.
npm Apache-2.0 ★ 4 -
Fast local judgment for AI agents: an MCP server of three small CPU models, no GPU or cloud
MIT -
s1-precogactive
Asymmetric AI delegation gateway pairing cloud reasoning with local GPU execution.
npm MIT -
Motor de memoria estruturada, governanca epistemologica e busca hibrida local integrado ao Obsidian via Model Context Protocol (MCP).
NOASSERTION ★ 1 -
Open Markdown Graph Base — local embedding server (transformers.js + gte-base), spoken to over stdio
npm MIT -
ensemble-mcpgithub.com/lynkbyte/ensembleupdated this year
MCP server providing vector memory, drift detection, model routing, skills discovery, session management, and codebase indexing for AI-assisted development pipelines.
pypi MIT License -
Zero-trust privacy layer, streaming PII redaction gateway & FastMCP server for AI coding agents
pypi MIT ★ 2 -
A desktop phone for your coding agent - it calls you over MCP and you have a natural, spoken back-and-forth. Voice runs fully on-device by default (sherpa-onnx: Parakeet STT + Kokoro TTS), with optional ElevenLabs. Built with Tauri v2 (Rust) + React.
MIT ★ 1 -
Local CPU OCR for scanned PDFs and images: searchable PDF, structured Markdown, Python API, and MCP. Windows beta.
Apache-2.0 ★ 1 -
robotmemgithub.com/robotmem/robotmemupdated this year
Robot Memory - Persistent memory system for AI robots. MCP Server + hybrid search + spatial retrieval.
cargo Apache-2.0 ★ 30 -
FastMCP 3.2 speech gateway: FunASR local STT, Gemini Live/TTS, Hume, ElevenLabs cloning, wake word, RAG. MCP server + React webapp.
MIT ★ 2 -
agentic-visiongithub.com/agentralabs/agentic-visionupdated this year
Persistent visual memory for AI agents — capture screenshots, embed with CLIP ViT-B/32, compare, recall. MCP server + Rust core library.
cargo MIT ★ 12 -
Veya is a local decision engine with typed choices, scores and yes/no decisions, retention-aware caching, local MCP tools and Python, TypeScript, Java and .NET integrations.
Apache-2.0 -
sup — run sup computer's tiny GPTs in your terminal, or serve them to an agent over MCP. Downloads a release's public artifacts and runs it on onnxruntime-node.
npm MIT -
A local semantic-recall MCP server for citation-integrity academic writing — fastembed ONNX embeddings on LanceDB, cross-lingual, fully offline. Recall-only by contract: fragments carry file and page; the source PDF supplies every quote. Companion to alexandria. 服務引文完整性寫作的本地語意召回 MCP 伺服器:跨語言、完全離線;只召回不引用,片段附檔案與頁碼,每一句引文都回原始 PDF 取。
MIT ★ 1 -
Local-only voice for Claude Code on macOS via Kyutai Pocket TTS — plugin + MCP server + /voice-mode skill
pypi MIT -
esia-mcpupdated this year
Servidor MCP local del detector de text generat per IA "és ia?" (català/castellà). Local MCP server for the "és ia?" on-device Catalan/Spanish AI-text detector.
npm MIT -
Store and recall robotic experiment data to improve task success by learning from past parameters, trajectories, and outcomes.
Apache-2.0 ★ 3 -
A self-hosted speech server with streaming transcription, speech synthesis, verified local model setup, WebSockets, HTTP, gRPC and MCP APIs.
Apache-2.0 ★ 3 -
A self-hosted embedding server with HTTP, gRPC and MCP APIs, verified model installation, batching, health checks, metrics and Windows and Linux releases.
Apache-2.0 ★ 5 -
Local memory + semantic search for AI agents. Indexes code and documents on your machine, queryable via MCP. Private, offline, fast. macOS arm64 + Linux x86_64.
★ 8 -
A self-hosted OCR server that turns PNG and JPEG images into structured text using local PP-OCRv5 models through HTTP, gRPC and MCP.
Apache-2.0 ★ 3 -
AI-assisted video editing framework built on MCP: 19 Python servers expose 45 tools for media analysis and preparation, enabling an AI agent to build frame-accurate DaVinci Resolve timelines.
MIT ★ 3 -
x402 payment facilitator for Stellar with a discovery layer: exact USDC settlement on Soroban with sponsored fees, automatic cataloging on settlement, semantic search with abstention, and an MCP server for agents. Live on Stellar testnet.
Apache-2.0 ★ 1 -
Ollama for vision: pull, run & serve computer-vision models (YOLO, SAM 2, Depth Anything, Grounding DINO) from your terminal. REST, WebSocket & MCP for AI agents and robots.
MIT ★ 1 -
Self-hosted OCR pipeline for PDFs, PPTX, and images with a native MCP server: 18% lower CER and 7.7x lower CPU latency vs. EasyOCR on a 14-page benchmark (ADR-documented). CLI + Streamlit UI + FastAPI, 677 automated tests.
MIT -
Eyes and ears for AI agents. Local, offline transcription, keyframes, on-screen text and a pre-publish review of any audio, video or image, as an MCP server, a CLI and a Node library. No Python, no cloud, no API key.
npm Apache-2.0 -
A session-scoped memory layer for LLMs, built on the Model Context Protocol. LatentContext gives AI assistants the ability to explicitly store and retrieve notes, decisions, and context within a single conversation — keeping the AI focused on the current task without cross-session data contamination.
npm MIT ★ 5 -
mcp-server-qdrantgithub.com/vaporif/mcp-server-qdrantupdated this year
Rust MCP server for Qdrant with local embeddings (Candle/ONNX)
cargo MIT -
vmafxgithub.com/VMAFx/vmafxactive
VMAFx: perceptual video quality assessment (VMAF), GPU-accelerated and SIMD-tuned. CUDA, SYCL, HIP and Metal backends checked against the CPU scores, AVX2/AVX-512/NEON, tiny-AI models, an MCP server, Kubernetes/Helm deployment. Fork of Netflix/vmaf; fork code EUPL-1.2, Netflix code BSD-2-Clause-Patent.
EUPL-1.2 ★ 7 -
Fully local MCP retrieval tool for coding agents. 161M encoder, ~1s per tool call, no API, no LLM in the query path. No file reads: line numbers, callers, callees and exact call sites, packed to a token budget. 2-4x the file coverage of BM25 and TF-IDF on SWE-Explore, and tops the paper's own sub-1B code model on CORE-Bench.
MIT ★ 5 -
VelociRAGgithub.com/HaseebKhalid1507/VelociRAGupdated this year
Lightning-fast RAG for AI agents. ONNX-powered, 4-layer fusion, MCP server. No PyTorch.
pypi MIT ★ 12