惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

J
Java Code Geeks
博客园 - 司徒正美
博客园 - 【当耐特】
爱范儿
爱范儿
OSCHINA 社区最新新闻
OSCHINA 社区最新新闻
IT之家
IT之家
人人都是产品经理
人人都是产品经理
雷峰网
雷峰网
酷 壳 – CoolShell
酷 壳 – CoolShell
freeCodeCamp Programming Tutorials: Python, JavaScript, Git & More
大猫的无限游戏
大猫的无限游戏
月光博客
月光博客
宝玉的分享
宝玉的分享
V
V2EX
S
SegmentFault 最新的问题
V
Visual Studio Blog
阮一峰的网络日志
阮一峰的网络日志
Martin Fowler
Martin Fowler
Jina AI
Jina AI
让小产品的独立变现更简单 - ezindie.com
让小产品的独立变现更简单 - ezindie.com
博客园_首页
L
LangChain Blog
D
Docker
腾讯CDC

Release notes from OmniClip-RAG

Release Caelune / 凯露恩 v0.5.6 · EllisMorrow/Caelune Release v0.5.5 · EllisMorrow/Caelune Release v0.5.4 · EllisMorrow/Caelune Release v0.5.2 · EllisMorrow/Caelune Release v0.5.1 · EllisMorrow/Caelune Release v0.5.0 · EllisMorrow/Caelune Release v0.4.13 · EllisMorrow/OmniClip-RAG Release v0.4.13 · EllisMorrow/Caelune Release v0.4.12 · EllisMorrow/OmniClip-RAG Release v0.4.12 · EllisMorrow/Caelune Release v0.4.9 · EllisMorrow/Caelune Release v0.4.11 · EllisMorrow/OmniClip-RAG Release v0.4.11 · EllisMorrow/Caelune Release v0.4.10 · EllisMorrow/OmniClip-RAG Release v0.4.10 · EllisMorrow/Caelune Release v0.4.9 · EllisMorrow/OmniClip-RAG Release v0.4.8 · msjsc001/OmniClip-RAG Release v0.4.8 · EllisMorrow/OmniClip-RAG Release v0.4.8 · EllisMorrow/Caelune Release v0.4.7 · msjsc001/OmniClip-RAG Release v0.4.7 · EllisMorrow/OmniClip-RAG Release v0.4.7 · EllisMorrow/Caelune Release v0.4.6 · msjsc001/OmniClip-RAG Release v0.4.6 · EllisMorrow/OmniClip-RAG Release v0.4.6 · EllisMorrow/Caelune Release v0.4.5 · msjsc001/OmniClip-RAG Release v0.4.5 · EllisMorrow/OmniClip-RAG Release v0.4.5 · EllisMorrow/Caelune Release v0.4.4 · msjsc001/OmniClip-RAG Release v0.4.4 · EllisMorrow/OmniClip-RAG
Release v0.4.0 · msjsc001/OmniClip-RAG
msjsc001 · 2026-03-18 · via Release notes from OmniClip-RAG

OmniClip RAG / 方寸引 v0.4.0

Summary

v0.4.0 turns OmniClip from a local desktop retrieval tool into a dual-shell product with a standard read-only MCP interface.

This release does not bolt MCP onto the GUI executable as a convenience flag. Instead, it introduces a second, headless shell over the same retrieval core:

  • OmniClipRAG.exe for desktop GUI
  • OmniClipRAG-MCP.exe for MCP stdio integration

The goal is to let MCP-capable AI clients use OmniClip's local knowledge retrieval safely, while preserving the existing GUI behavior and keeping the backend single-sourced.

Highlights

  • added a dedicated headless MCP executable instead of reusing the windowed GUI build
  • kept MCP V1 intentionally small and stable with only two read-only tools:
    • omniclip.status
    • omniclip.search
  • introduced a shared headless bootstrap so GUI and MCP now align on the same RuntimeContext, DataPaths, and QueryService startup path
  • made MCP search explicitly degradation-aware
    • semantic path available -> hybrid
    • semantic path unavailable -> lexical_only
    • warnings are surfaced instead of hidden
  • documented setup and client examples for Claude Desktop, Cursor, and Cline

Included In This Release

Shared core

  • added a reusable headless bootstrap layer
  • extracted launcher bootstrapping into shared helpers
  • kept GUI and MCP as separate shells over the same retrieval kernel

MCP server

  • added omniclip_rag/mcp/core.py
  • added omniclip_rag/app_entry/mcp.py
  • added launcher_mcp.py
  • added omniclip.status
  • added omniclip.search
  • added --mcp-selfcheck

Packaging

  • added OmniClipRAG-MCP.spec
  • updated build.py to produce both:
    • OmniClipRAG-v0.4.0/
    • OmniClipRAG-MCP-v0.4.0/
  • kept both packages lean and free of runtime payloads, model caches, and user data

Documentation

  • added MCP_SETUP.md
  • added examples/mcp/*.json
  • added plain-language README onboarding, including a tested Jan.ai stdio configuration example
  • added the in-repo execution record:
    • plans/OmniClip RAG MCP接入实施计划.md
  • updated README, README.zh-CN, ARCHITECTURE, and CHANGELOG for the MCP line

V1 Boundary

This release intentionally does not include:

  • write operations
  • build/delete/config mutation through MCP
  • prompts/resources
  • Streamable HTTP
  • remote auth / multi-tenant design

The first goal is a durable, read-only, local MCP interface over the existing retrieval core.