kronode 0.2.0

Extract team conventions from git history, serve via MCP to any AI coding tool

Extract team conventions from git history. Serve them to any AI coding tool via MCP.

Quick Start · Setup AI Tool · Modes · MCP Tools

Three commands. Zero API keys. Works with Claude Code, Cursor, Codex, Copilot, and 10+ more.

pip install kronode
kronode init .
kronode ingest
kronode serve

---

What it does

Kronode reads your local git repository and extracts:

• Conventions — commit message patterns, team naming rules, directory structure
• Documentation — README, docs/, architecture decisions from markdown files
• PR review knowledge — what reviewers actually teach in code reviews (optional)

Then serves it all via MCP to any AI coding tool — so your AI assistant knows your team's patterns, not just generic best practices.

Quick Start

# Install
pip install kronode

# Point to your repo
cd ~/projects/your-repo
kronode init .

# Extract conventions + docs from git history
kronode ingest

# Optional: also fetch PR review comments (uses gh CLI auth)
kronode ingest --with-prs

# Start MCP server
kronode serve

Then connect your AI tool (see Setup below).

How it works

Your Git Repo               Kronode                    AI Coding Tool
┌──────────────┐     ┌──────────────────┐     ┌──────────────────┐
│ git log      │────▸│ Convention       │     │ Claude Code      │
│ git blame    │     │ Extraction       │     │ Cursor           │
│ docs/*.md    │────▸│                  │◂───▸│ Codex            │
│ PR reviews   │────▸│ SQLite DB        │ MCP │ Copilot          │
│ (optional)   │     │ Local Embeddings │     │ Windsurf         │
└──────────────┘     └──────────────────┘     │ + 7 more         │
                                               └──────────────────┘

Everything runs locally. No data leaves your machine unless you opt into BYOK mode.

CLI Reference

Command	Description
kronode init .	Initialize for a local repo
kronode ingest	Extract conventions + docs from git history
kronode ingest --with-prs	Also fetch PR review comments (uses gh CLI or --token)
kronode serve	Start MCP server (stdio)
kronode serve --http	Start MCP server (HTTP/SSE for Cursor, remote)
kronode query "task" -f file.py	Test context retrieval from terminal
kronode setup <tool>	Print MCP config for an AI tool
kronode status	Show convention/doc counts
kronode add "convention rule"	Manually add a convention
kronode purge	Delete all ingested data

Setup Your AI Tool

Claude Code

claude mcp add kronode -- kronode serve

Cursor

Add to .cursor/mcp.json:

{
  "mcpServers": {
    "kronode": {
      "command": "kronode",
      "args": ["serve"]
    }
  }
}

OpenAI Codex

Add to ~/.codex/config.toml:

[mcp_servers.kronode]
command = "kronode"
args = ["serve"]

[mcp_servers.kronode.tools.get_context]
approval_mode = "approve"

[mcp_servers.kronode.tools.get_doc]
approval_mode = "approve"

GitHub Copilot (VS Code)

Add to .vscode/mcp.json:

{
  "servers": {
    "kronode": {
      "command": "kronode",
      "args": ["serve"]
    }
  }
}

Run kronode setup --list to see all supported tools.

Modes

Local Mode (default)

Zero API keys, zero network. Reads from your .git/ directory.

• Conventions from commit message patterns + git structure
• Documentation from markdown files
• Local embeddings via fastembed (ONNX, ~50MB)

Local + PRs

Add --with-prs to also fetch PR review comments. If gh CLI is installed and authenticated, no token needed. Otherwise pass --token <PAT>.

# With gh CLI (recommended)
gh auth login
kronode ingest --with-prs

# Or with explicit PAT
kronode ingest --with-prs --token ghp_xxx

BYOK Mode

Set OPENAI_API_KEY for richer LLM-based convention extraction from PR diffs.

OPENAI_API_KEY=sk-xxx kronode ingest --with-prs

Configuration

Copy .env.example to ~/.kronode/.env:

# Number of PRs to analyze (default: 200)
KRONODE_PR_COUNT=200

# Minimum confidence threshold (default: 0.5)
KRONODE_MIN_CONFIDENCE=0.5

# Max conventions per query (default: 20)
KRONODE_MAX_CONVENTIONS=20

MCP Tools

Kronode exposes 2 MCP tools:

get_context

Returns everything an AI needs before coding:

• Top conventions ranked by file-level relevance (with match reasons)
• Related documentation from repo markdown
• Action plan summary
• Directly and loosely related issues (if indexed)

get_doc

Returns the full content of a documentation page when get_context returned a truncated snippet.

Tech Stack

• Python 3.11+ — async throughout
• SQLite — zero-config local database (PostgreSQL optional)
• fastembed — ONNX local embeddings (~50MB, not 2GB PyTorch)
• MCP — Model Context Protocol for AI tool integration
• click + rich — CLI framework

Development

git clone https://github.com/GauravGill90/kronode.git
cd kronode
uv venv .venv && source .venv/bin/activate
uv pip install -e ".[byok]"
kronode --help

License

MIT