"ONNX" matching MCP servers:

Local RAG system for Claude Code with hybrid search (semantic + BM25), cross-encoder reranking, markdown-aware chunking, and 12 MCP tools. Zero external servers, pure ONNX in-process.

Local semantic recall MCP server for citation-integrity workflows. Enables semantic search over a text corpus, returning file paths and page numbers for verification, with 100% local processing using ONNX embeddings and LanceDB.

Semantix-Verify is an MCP server for semantic validation of AI/LLM outputs. It exposes a single tool, verify_text_intent(text, intent_description, threshold), which uses a local quantized NLI cross-encoder (INT8 ONNX) to return a 0.0–1.0 probability that the text satisfies the given intent — and, when it doesn't, a structured correction prompt for agent retry loops. Useful for building com

Knowledge graph for token-efficient code reviews. Builds a structural map of your codebase with Tree-sitter, tracks changes incrementally, and gives AI agents precise context via MCP tools. Features fixed multi-word search, qualified call resolution, dual-mode embedding (ONNX local + LiteLLM cloud), and output pagination.

Semantic vector DB as an MCP server for Claude Code — SQLite + FTS5 + local ONNX embeddings. Gives Claude persistent, searchable project knowledge across conversations.

The open retrieval layer for AI agents. Index your entire project — code, docs, legal, research, data — and serve surgical context via MCP. FTS5 full-text search, optional semantic search (FastEmbed/ONNX), 10 built-in parsers, incremental auto-sync.

A drop-in replacement for Anthropic's memory server that utilizes SQLite to ensure data integrity and concurrent access. It enhances the original functionality with semantic search capabilities using vector embeddings and ONNX models.

Semantic memory server for AI agent teams. Stores, searches, and retrieves knowledge across sessions using pluggable vector backends with local ONNX embeddings, exposed as an MCP server.

Build a paid MCP server that charges AI agents per call in USDC, with automatic payment handling via HTTP 402 and EIP-3009.

Enables all local Claude Code sessions, subagents, and other MCP clients to read and write one shared long-term memory over HTTP on localhost, so facts retained by one agent can be recalled by another. It extracts atomic facts, entities, and observations into an embedded local Postgres with ONNX embeddings and reranking, requiring no API key.

Local MCP server for neural text-to-speech using Kokoro ONNX engine on CPU, supporting SSML tags, multiple voice profiles, and zero-GPU operation for low-latency speech synthesis.

FaceTron is a high-performance face embedding server with ONNX Runtime, dynamic multi-model loading, and MCP-compatible OpenAPI metadata for offline face recognition and detection.

An MCP server that autonomously optimizes ONNX ML models for Arm64 deployment, providing tools to analyze models, apply real INT8 quantization, benchmark performance, and generate Arm64-optimized Docker deployment packages.

Enables AI assistants to estimate monocular depth maps from local or remote images and frame-by-frame video using bundled ONNX Depth Anything V2 models, without requiring PyTorch. Also exposes tools for listing and pre-downloading checkpoints and checking the runtime backend, with automatic GPU/CPU selection.

A Flask-based MCP server that connects to AutoCAD and executes AutoLISP code to enable architectural floor plan creation and management through a REST API.