Ctxnest

Name: Ctxnest
Author: safiyu

@safiyu

Ctxnest について

Local Brain/Hands/Eyes loop for AI coding agents. Persistent markdown knowledge vault with FTS5 search, sandboxed command execution via project-defined tools, and a feedback journal that compounds every session. 46 MCP tools. 100% local, 0% egress.

#mcp #ai #claude-code #context #memory #local #developer-tools

基本情報

カテゴリ

その他

トランスポート

stdio

公開者

safiyu

投稿者

Safiyu Vattaparambil

設定

以下の設定を使って、このサーバーを MCP 対応クライアントに追加してください。

{
  "mcpServers": {
    "ctxnest": {
      "command": "npx",
      "args": [
        "-y",
        "ctxnest-mcp"
      ]
    }
  }
}

ツール

ツールは検出されませんでした

ツールは README から自動的に抽出されます。メンテナーは ## Tools という見出しの下に記載することで、このタブに反映できます。

概要

What is CtxNest?

CtxNest is a local MCP server that gives AI coding agents a persistent knowledge vault (Brain), sandboxed command execution (Hands), and learning feedback (Eyes). It integrates with local markdown files, SQLite FTS5, and git, and is designed for developers using Claude Code, Cursor, Gemini, or Codex.

How to use CtxNest?

Configure commands in a ctxnest.json file; the server then exposes them as sandboxed MCP tools. The agent can invoke any of the 46 MCP tools directly, with optional human approval before execution. The server runs entirely locally with no external network calls.

Key features of CtxNest

Markdown knowledge vault with SQLite FTS5 full-text search
Git-backed versioning and cross-project context retrieval
Sandboxed tool execution: no shell, locked working directory, clean environment
Optional human approval gate before any command runs
Stdout capture, Brain diffs, and journal_append for agent learning
100 % local, zero egress of data

Use cases of CtxNest

An AI agent reads project documentation from the vault instead of manual copy-pasting
A developer defines safe shell commands in ctxnest.json and lets the agent run them
The agent records learnings after each task so the next session starts smarter
Cross-project context sharing across multiple codebases without duplicating docs

FAQ from CtxNest

How does CtxNest keep command execution safe?

Commands are declared in ctxnest.json and exposed as sandboxed MCP tools with no shell access, a locked working directory, a clean environment, and optional human approval before anything runs. The agent can only do what you explicitly declared.

What data leaves my machine?

None. CtxNest is 100 % local and 0 % egress. Your data never leaves your SSD.

Which AI coding agents does CtxNest work with?

It works with Claude Code, Cursor, Gemini, and Codex.

How many MCP tools does CtxNest provide?

CtxNest exposes 46 MCP tools covering knowledge retrieval, command execution, and journaling.

How does CtxNest persist what the agent learns?

It captures stdout, produces Brain diffs, and supports journal_append so the agent writes what it learned back into the vault. Next sessions start smarter than the last.