"SEPA" matching MCP servers:

Enables AI agents and software to detect ambiguous wording in text and receive location, category, explanation, and suggested clarification questions for each signal.

Enables users to analyze already-persisted single-patent evaluations to discover cross-patent batch issues such as language, technical-domain, input-structure sensitivity, and language-structure interactions, without modifying existing MCP, ledger, or single-record evaluation JSON.

Content-addressed vocabulary protocol: 452 cognitive patterns with cryptographic identity. Agents sharing a handle (e.g. StateLock#5602) provably share meaning — mismatched vocabularies halt rather than silently drift.

Implements the Genetic-Evolutionary Prompt Architecture for automatic prompt optimization, providing tools to optimize, quickly improve, and contextually adapt prompts.

Audits KiCad PCB designs against South-East Asian fab capabilities using an IPC-compatible DFM rules engine with optional AI review via Ollama.

Enables audio source separation into vocals and instruments using ComfyUI MelBandRoFormer models via a set of MCP tools.

Universal MCP server for the full Replicate catalog: image, video, audio, music, speech (TTS+STT), LLMs, vision, upscaling, segmentation, embeddings, voice cloning, 3D, and lipsync. 36 tools and 63 curated models, with async batch + DAG pipeline orchestration and auto-download of outputs.

Enables a private, local-first AI agent that uses your Obsidian vault as memory, exposing an ask_seastar tool to any MCP host and consuming other MCP servers while running entirely on your machine with any OpenAI-compatible local model.