FK

fandry96/k3-mcp-toolbox-public

Developer tools

このページは AI により自動翻訳されています。技術的な詳細は原文をご確認ください。 原文を見る

11 stars 4 forks 品質 84 トレンド 38EPL-2.0

Welcome to the open-source release of the and . These are the production-tested tools and cognitive primitives we use to power autonomous coding agents on Windows and Linux, open-sourced for the...

概要

Welcome to the open-source release of the and . These are the production-tested tools and cognitive primitives we use to power autonomous coding agents on Windows and Linux, open-sourced for the agentic AI community. solves the 5 most frustrating bottlenecks encountered during long-running coding agent sessions: 1. : Queries using Matryoshka Representation Learning (MRL), pre-cached normalized matrix dot products, and $O(N)$ candidate selection via np.argpartition. 2. : Surgically reaps orphaned test browsers (chromedriver, msedgedriver) and frees locked dev ports (3000, 8080, etc.) without killing your active IDE or agent sessions. 3. : Automatically spins up and merges isolated Git worktrees (.worktrees/ ) with strict binary allowlists and automatic conflict rollbacks, preventing parallel agents from corrupting git staging. 4.

README

以下は原文 README の抜粋です。原文を見る。

K3 MCP Toolbox & Antigravity Logicware

“Production Middleware for the Agentic Era.”

Welcome to the open-source release of the K3 MCP Toolbox and Antigravity Logicware. These are the production-tested tools and cognitive primitives we use to power autonomous coding agents on Windows and Linux, open-sourced for the agentic AI community.


⚡ TL;DR

K3 MCP Toolbox solves the 5 most frustrating bottlenecks encountered during long-running coding agent sessions:

  1. Blazing-Fast Local Semantic Memory (k3-mrl-memory): Queries 25,300+ vectors in 0.22 seconds on local CPU using Matryoshka Representation Learning (MRL), pre-cached normalized matrix dot products, and $O(N)$ candidate selection via np.argpartition.
  2. OS & Runtime Sentinel (k3-agent-ops): Surgically reaps orphaned test browsers (chromedriver, msedgedriver) and frees locked dev ports (3000, 8080, etc.) without killing your active IDE or agent sessions.
  3. Subagent Git Worktree Isolation (k3-worktree-ops): Automatically spins up and merges isolated Git worktrees (.worktrees/) with strict binary allowlists and automatic conflict rollbacks, preventing parallel agents from corrupting git staging.
  4. Secure Document Intelligence (k3-doc-intel): Ingests and chunks PDFs and DOCX files with a custom _RestrictedUnpickler whitelist that blocks arbitrary code execution vulnerabilities on untrusted files.
  5. Local LLM Lifecycle (k3-local-llm): Manages bundled llama-server.exe and local GGUF models for offline embeddings and completions with automated child process reaping.
  6. Gemini 3.8 Flash GA Primitives (antigravity-logicware): Config sanitizers that strip banned legacy sampling parameters (temperature, top_p, penalties), enforce strict 1:1 FunctionCall.id matching, and provide structured update(previous_step, plan, next_step) progress declarations.

🤖 AI Agent Setup Guide (When an AI Pulls This Repo)

If you are an AI assistant or coding agent (Cursor, Claude Desktop, Antigravity, Cline, Windsurf, Aider, or custom ADK loop) asked to set up or use this repository, follow these precise steps:

1. Install Dependencies

# From repository root:
pip install -r k3-mcp-toolbox/requirements.txt
pip install -e antigravity-logicware

2. Verify Everything (100% Offline Self-Test)

Run both test suites before touching code:

# Test the 5-server FastMCP fleet (27 assertions):
GitHub で README 全文を見る

インストール

This server does not publish a one-line install command.

Open the repository installation guide MCP 初心者?mcpskill.org のガイドから始めましょう →

設定

{ "mcpServers": { "k3-agent-ops": { "command": "python", "args": ["<ABSOLUTE_PATH>/k3-mcp-toolbox/servers/k3_agent_ops.py"] }, "k3-mrl-memory": { "command": "python", "args": ["<ABSOLUTE_PATH>/k3-mcp-toolbox/servers/k3_mrl_memory.py"], "env": { "GEMINI_API_KEY": "optional_for_live_embedding_generation", "MRL_INDEX_PATH": "<ABSOLUTE_PATH>/k3-mcp-toolbox/src/mrl_index.pkl" } }, "k3-worktree-ops": { "command": "python", "args": ["<ABSOLUTE_PATH>/k3-mcp-toolbox/servers/k3_worktree_ops.py"] }, "k3-doc-intel": { "command": "python", "args": ["<ABSOLUTE_PATH>/k3-mcp-toolbox/servers/k3_doc_intel.py"] }, "k3-local-llm": { "command": "python", "args": ["<ABSOLUTE_PATH>/k3-mcp-toolbox/servers/k3_local_llm.py"] } } }