ktme-cli

ktme - Knowledge Transfer Me

A Rust-based CLI tool and MCP server for automated documentation generation from code changes. Integrates with AI coding assistants like Claude Code, Cursor, Windsurf, and other MCP-compatible clients.

Table of Contents

• Overview
• Core Concepts
• Features
• Installation
• AI Agent Integration
• Quick Start
• Configuration
• MCP Tools
• Knowledge Search (RAG)
• Usage Examples
• Architecture
• Contributing
• License

Overview

ktme (Knowledge Transfer Me) is a CLI tool and MCP server that bridges code changes with documentation. It uses the Model Context Protocol (MCP) to communicate with AI coding assistants, enabling them to:

• Read and analyze code changes from Git
• Generate documentation based on diffs
• Update existing docs intelligently
• Publish to Markdown files or Confluence

graph LR
    subgraph AI Assistants
        Claude[Claude Code]
        Cursor[Cursor]
        Windsurf[Windsurf]
    end

    subgraph ktme MCP Server
        Server[MCP Server]
        Tools[Tools]
    end

    Git[Git Repository] --> Tools
    Claude --> Server
    Cursor --> Server
    Windsurf --> Server
    Server --> Tools
    Tools --> Docs[Documentation]

How It Works

• You configure ktme as an MCP server in your AI assistant (Claude Code, Cursor, etc.)
• AI assistant connects to ktme via the Model Context Protocol
• AI uses ktme tools to read code changes, query mappings, and generate docs
• Documentation is created in your preferred format (Markdown, Confluence)

Core Concepts

Multi-Source Change Extraction

ktme can extract code changes from multiple sources:

# From commits
ktme extract --commit abc123

# From staged changes
ktme extract --staged

# From pull requests
ktme extract --pr 456 --provider github

# From commit ranges
ktme extract --commit v1.0.0..v1.1.0

Service-to-Documentation Mapping

Map your services to their documentation locations:

# ~/.config/ktme/mappings.toml
[[services]]
name = "api-gateway"
path = "projects/api-gateway"
docs = [
  { type = "markdown", location = "projects/api-gateway/README.md" },
  { type = "confluence", location = "https://company.atlassian.net/wiki/display/API/APIGateway" }
]

MCP Integration

ktme exposes tools through the Model Context Protocol, enabling AI assistants to:

• Read extracted code changes
• Query service-document mappings
• Generate documentation based on changes
• Update existing documentation intelligently

Features

Feature	Status	Description
CLI Interface	✅ IMPLEMENTED	Complete command structure with extract, generate, mapping, MCP, and config commands
Git Diff Extraction	✅ IMPLEMENTED	Extract from commits and staged changes (PR extraction not yet implemented)
AI-Powered Generation	✅ IMPLEMENTED	Generate documentation using OpenAI or Claude AI models
Service Mapping Storage	✅ IMPLEMENTED	SQLite database for storing service-to-documentation mappings
MCP Server with Tools	✅ IMPLEMENTED	Functional MCP server with tools for reading changes, mappings, and generating docs
Configuration Management	✅ IMPLEMENTED	TOML-based configuration with environment variable support
Database Layer	✅ IMPLEMENTED	Complete SQLite models and repositories for all features
JSON Output Format	✅ IMPLEMENTED	Export diffs and data in JSON format
Basic Error Handling	✅ IMPLEMENTED	Comprehensive error types and propagation
Git Platform Support	🔄 PARTIAL	Local Git ✅, GitHub/GitLab PR support framework exists
Template System	🔄 FRAMEWORK	Template structure in place, basic rendering implemented
Documentation Providers	🔄 BASIC	Markdown file writer ✅, Confluence provider structure exists
PR Extraction	❌ NOT IMPLEMENTED	GitHub/GitLab API integration for PR diffs
Knowledge Search/RAG	❌ NOT IMPLEMENTED	Vector embeddings and semantic search features
Confluence Sync	❌ NOT IMPLEMENTED	No Confluence API integration implemented
Feature Mapping System	❌ NOT IMPLEMENTED	Feature-to-document mapping not implemented

Current Implementation Status

✅ Fully Implemented

• CLI Interface: Complete command-line interface with all major commands (extract, generate, mapping, MCP, config)
• Git Operations: Functional diff extraction from commits and staged changes using git2-rs
• AI Integration: Working OpenAI and Claude API clients with automatic provider detection
• MCP Server: Fully functional MCP server with STDIO transport and working tools
• SQLite Storage: Complete database layer with models and repositories
• Service Mapping: Add, list, get, remove service-to-document mappings
• Configuration: TOML-based configuration management with environment variables
• Error Handling: Comprehensive error types and proper propagation

🔄 Partially Implemented

• Documentation Generation: Basic template-based generation from Git diffs
• Markdown Writer: Simple file-based documentation output
• PR Extraction: Framework exists but GitHub/GitLab API integration not implemented
• Confluence Provider: Structure in place but API calls not implemented
• Template System: Basic template structure exists
• HTTP Transport: SSE mode structure implemented but not tested

❌ Not Yet Implemented

• Knowledge Search/RAG: No vector embeddings, semantic search, or RAG functionality
• Confluence Sync: No incremental sync or API integration
• Feature Mapping: No feature-to-document mapping system
• Bitbucket Integration: Provider structure not implemented
• Advanced Templates: No sophisticated template rendering
• Web UI: No web interface for managing mappings

⚠️ Known Issues

• 103 compilation warnings (mostly unused code - expected for incomplete features)
• PR extraction returns error - GitHub/GitLab API not implemented
• Some MCP tools have basic implementations (generate_documentation creates simple markdown)
• Confluence provider structure exists but API calls not implemented

Installation

Option 1: Install from NPM (Recommended)

npm install -g ktme-cli

# Verify installation
ktme --version

Option 2: Install from Source

Prerequisites:

• Rust 1.70 or higher
• Git 2.30 or higher
• Access to an AI model API (Claude, GPT, etc.)

Build from source:

git clone https://github.com/FreePeak/ktme.git
cd ktme
cargo build --release

Using Cargo:

cargo install --path .

Development Workflow

For active development, use one of these methods to update your ktme command after code changes:

Method 1: Makefile (Recommended)

# Quick development cycle
make dev

# Or step by step
make build-release
make install-dev

Method 2: Rebuild and Replace

# After code changes, rebuild
cargo build --release

# Copy to system location (one-time setup)
sudo cp ./target/release/ktme /usr/local/bin/ktme

Method 3: Cargo Install (Easiest)

# After code changes, reinstall globally
cargo install --path --force .

Method 4: Development Alias Add to your ~/.zshrc or ~/.bashrc:

alias ktme="cd /path/to/ktme && ./target/release/ktme"

Method 5: PATH Update Add to your shell config:

export PATH="/path/to/ktme/target/release:$PATH"

Using Makefile

The project includes a Makefile for common development tasks:

# Show all available targets
make help

# Build and install in development mode
make dev

# Build in release mode
make build-release

# Run tests and formatting
make ci

# MCP server management
make run-mcp    # Start server
make stop-mcp   # Stop server
make status-mcp # Check status

Verify Installation

ktme --version

Note: After installing ktme-cli from NPM, the command available is ktme (without -cli).

AI Agent Integration

ktme uses the Rust MCP SDK to implement the Model Context Protocol, enabling seamless communication with AI coding assistants.

Claude Code

Add to your Claude Code MCP settings (~/.claude/mcp_settings.json):

{
  "mcpServers": {
    "ktme": {
      "command": "ktme",
      "args": ["mcp", "start"],
      "env": {
        "KTME_MCP_API_KEY": "your-api-key"
      }
    }
  }
}

Cursor

Add to Cursor's MCP configuration (.cursor/mcp.json):

{
  "servers": {
    "ktme": {
      "command": "ktme",
      "args": ["mcp", "start", "--stdio"]
    }
  }
}

Windsurf

Add to Windsurf's MCP settings:

{
  "mcp": {
    "servers": {
      "ktme": {
        "command": "ktme",
        "args": ["mcp", "start", "--stdio"]
      }
    }
  }
}

SSE Mode (HTTP)

For HTTP-based connections, start ktme in SSE mode:

ktme mcp start --sse --host 127.0.0.1 --port 8080

Then configure your AI assistant to connect to http://127.0.0.1:8080.

Quick Start

1. Install ktme

npm install -g ktme-cli

2. Initialize Configuration

ktme config init

3. Set Up API Key

export KTME_MCP_API_KEY="your-api-key"

4. Map Your Service

ktme mapping add user-api --file ~/projects/user-api/README.md

5. Generate Documentation

cd ~/projects/user-api
ktme generate --commit HEAD --service user-api

Important: The NPM package is named ktme-cli, but after installation, you use the ktme command.

Configuration

Configuration File

Location: ~/.config/ktme/config.toml

[general]
default_directory = "~/projects"
log_level = "info"

[git]
default_branch = "main"
include_merge_commits = false

[mcp]
model = "claude-3-5-sonnet-20241022"
max_tokens = 4096
temperature = 0.7

[documentation]
default_format = "markdown"
template_directory = "~/.config/ktme/templates"

[confluence]
base_url = "https://your-company.atlassian.net"
auth_type = "token"
space_key = "DOCS"

Environment Variables

Variable	Description
KTME_MCP_API_KEY	AI model API key
KTME_MCP_MODEL	Model identifier
CONFLUENCE_API_TOKEN	Confluence authentication token
CONFLUENCE_USERNAME	Confluence username
KTME_LOG_LEVEL	Logging level (debug, info, warn, error)
OPENAI_API_KEY	OpenAI API key for embeddings (optional)

Knowledge Search Configuration

[knowledge]
# Cache location (default: ~/.config/ktme/knowledge.db)
cache_path = "~/.config/ktme/knowledge.db"

# Embedding provider: "openai" or "local"
embedding_provider = "local"

# Confluence spaces to sync
sync_spaces = ["DEV", "API", "DOCS"]

# Auto-sync interval in hours (0 = manual only)
auto_sync_interval = 0

[knowledge.chunking]
# Chunk size for RAG (tokens)
chunk_size = 512
chunk_overlap = 50

MCP Tools

ktme exposes the following tools through MCP:

Documentation Tools

Tool	Description
read_changes	Read extracted diff from a file or Git
generate_documentation	Create new documentation from changes
update_documentation	Modify existing documentation

Mapping Tools

Tool	Description
get_service_mapping	Retrieve document URL for a service
list_services	List all mapped services
add_mapping	Add a new service-to-doc mapping

Git Tools

Tool	Description
extract_commit	Extract changes from a specific commit
extract_pr	Extract changes from a pull request
list_commits	List commits in a range

Knowledge Search Tools

Tool	Description
search_knowledge	Search documentation using natural language queries
get_document	Retrieve full document content by ID or URL
list_documents	List documents filtered by team, tags, or source
sync_documents	Trigger sync from Confluence or other sources

Feature Mapping Tools

Tool	Description
get_feature	Get a feature/screen with all related documentation
map_feature_document	Link a feature to a documentation page
list_features	List all features filtered by team

Knowledge Search (RAG) - NOT IMPLEMENTED

⚠️ WARNING: This section describes planned features that are not yet implemented. The knowledge search, RAG capabilities, and related tools do not exist in the current codebase.

ktme is designed to include a powerful knowledge search system that will enable teams to search documentation across Confluence and local files using natural language queries through MCP-connected AI assistants.

Planned Architecture

graph TB
    subgraph "Users (via Cursor/Claude)"
        Backend[Backend Team]
        Mobile[Mobile Team]
    end

    subgraph "MCP Server"
        SearchTool[search_knowledge]
        GetDocTool[get_document]
        SyncTool[sync_documents]
    end

    subgraph "Local Cache"
        SQLite[(SQLite)]
        FTS[FTS5 Index]
        Embeddings[Embeddings]
    end

    subgraph "Source of Truth"
        Confluence[(Confluence)]
    end

    Backend -.-> SearchTool
    Mobile -.-> SearchTool
    SearchTool -.-> FTS
    SearchTool -.-> Embeddings
    GetDocTool -.-> SQLite
    Confluence -.->|"Incremental Sync"| SQLite
    SQLite -.-> FTS
    SQLite -.-> Embeddings

How It Will Work

• Confluence is the source of truth - All documentation lives in Confluence
• Local SQLite cache - Documents will be synced to a local cache for fast searching
• Hybrid search - Will combine keyword matching (FTS5) with semantic search (embeddings)
• Incremental sync - Will only fetch documents modified since last sync

Planned Knowledge Search Tools

Tool	Status	Description
search_knowledge	❌ Not Implemented	Search documentation using natural language
get_document	❌ Not Implemented	Retrieve full document content by ID
list_documents	❌ Not Implemented	List documents by team, tags, or source
sync_documents	❌ Not Implemented	Sync documents from Confluence

Syncing Documents

# Initial full sync from Confluence
ktme sync --space DEV --full
ktme sync --space API --full

# Incremental sync (only changed documents)
ktme sync --space DEV

# Sync all configured spaces
ktme sync --all

Searching Knowledge

# CLI search
ktme search "user dashboard overview"
ktme search "payment integration" --team mobile
ktme search "authentication flow" --tag "feature:auth"

# Via MCP (AI assistant uses these tools)
# search_knowledge("user dashboard overview")
# search_knowledge("how does the API gateway work", team="backend")

Tagging Documents

Documents can be tagged for better organization:

# Add tags to indexed documents
ktme tag DOC_ID --team mobile --tag "screen:user_dashboard"
ktme tag DOC_ID --tag "feature:main_screen"

# Search by tags
ktme search --tag "screen:*" --team mobile

Cache Location

The knowledge cache is stored locally per user:

~/.config/ktme/
  config.toml           # Configuration
  ktme.db               # Service mappings
  knowledge.db          # Knowledge search cache
    documents           # Cached document content
    documents_fts       # Full-text search index
    document_chunks     # RAG chunks for context
    embeddings          # Vector embeddings
    sync_state          # Last sync timestamps

The cache is ephemeral and can be regenerated by re-syncing from Confluence.

Feature Mapping - NOT IMPLEMENTED

⚠️ WARNING: This section describes planned features that are not yet implemented. The feature mapping system does not exist in the current codebase.

Features are planned to represent logical units like mobile screens, components, or business flows. They will be mapped to documentation and services.

Planned Commands:

# Add a feature (mobile screen)
ktme feature add user_dashboard \
    --team mobile \
    --display-name "User Dashboard Screen" \
    --aliases "dashboard,main screen,overview"

# Map feature to documentation
ktme feature map user_dashboard \
    --doc-url "https://confluence.company.com/display/MOBILE/User+Dashboard"

# Link feature to backend service
ktme feature link user_dashboard --service api-gateway

# Get feature with all related docs
ktme feature get user_dashboard

Planned Feature-Service-Document Relationships:

Feature (Mobile)          Service (Backend)       Document (Confluence)
----------------          -----------------       ---------------------
user_dashboard      <---> api-gateway       ---> "User Dashboard Design Doc"
                    <---> user-service      ---> "User API Reference"
                                            ---> "Authentication Flow"

Example: Mobile Team Searching Documentation

Mobile Dev in Cursor: "Find docs about the user dashboard screen"

AI Assistant calls: search_knowledge("user dashboard screen")

System searches:
1. Features FTS: Matches "user_dashboard" via alias "dashboard"
2. Documents FTS: Matches documents containing "user interface"
3. Returns merged results with feature->document mappings

Results returned:
1. "User Dashboard - Main Interface"
   URL: https://confluence.company.com/display/MOBILE/User+Dashboard
   Team: mobile
   Feature: user_dashboard
   Related Services: [api-gateway, user-service]
   Summary: "The user dashboard displays account information and recent activity..."

2. "API Gateway Documentation"
   URL: https://confluence.company.com/display/BACKEND/API+Gateway
   Team: backend
   Service: api-gateway
   Related Features: [user_dashboard, authentication]
   Summary: "Central gateway for routing API requests to backend services..."

Usage Examples

⚠️ NOTE: The examples below show the intended usage, but most CLI commands are not yet implemented. The project compiles but command handlers are missing.

Document a Feature Branch (Planned)

# Extract changes from feature branch
ktme extract --commit main..feature/new-auth --output /tmp/feature.json

# Generate documentation
ktme generate --input /tmp/feature.json --service user-service --type api-doc

Update Changelog from PR (Planned)

# Extract PR changes
ktme extract --pr 456 --provider github

# Update changelog section
ktme update --pr 456 --service api-gateway --section "Changelog"

Publish to Confluence (Planned)

# Configure Confluence
export CONFLUENCE_API_TOKEN="your-token"
ktme config set confluence.base_url "https://company.atlassian.net"

# Map and generate
ktme mapping add user-service --url "https://company.atlassian.net/wiki/spaces/DOCS/pages/12345"
ktme generate --commit abc123 --service user-service --format confluence

Document Staged Changes (Planned)

# Stage your changes
git add src/main.rs

# Generate documentation before committing
ktme generate --staged --service my-service

# Commit both code and docs
git add .
git commit -m "feat: Add new feature with documentation"

What Actually Works Now

# Build the project
cargo build --release

# Extract changes from staging
./target/release/ktme extract --staged --output changes.json

# Extract changes from a commit
./target/release/ktme extract --commit HEAD --output last-commit.json

# Start MCP server with STDIO transport
./target/release/ktme mcp start --stdio

# Check version
./target/release/ktme --version

# The binary compiles and runs with working Git operations

Architecture

graph TB
    subgraph CLI Layer
        CLI[CLI Interface]
        Config[Config Manager]
    end

    subgraph Core Layer
        Git[Git Reader]
        Extractor[Diff Extractor]
        Generator[Doc Generator]
        Storage[Mapping Storage]
    end

    subgraph MCP Layer
        Server[MCP Server]
        Tools[MCP Tools]
    end

    subgraph Output Layer
        MD[Markdown Writer]
        Confluence[Confluence Writer]
    end

    CLI --> Git
    CLI --> Config
    Config --> Storage
    Git --> Extractor
    Extractor --> Server
    Server --> Tools
    Tools --> Generator
    Storage --> Generator
    Generator --> MD
    Generator --> Confluence

Component Overview

Component	Purpose
CLI Interface	Command parsing and user interaction
Git Reader	Extract changes from Git repositories
Diff Extractor	Parse and structure Git diffs
MCP Server	Model Context Protocol communication
Doc Generator	Transform AI output to documentation
Mapping Storage	Service-to-document mappings
Writers	Output to Markdown or Confluence

Supported Platforms

Platform	PR Extraction	Commit Extraction
GitHub	Yes	Yes
GitLab	Yes	Yes
Bitbucket	Yes	Yes
Local Git	N/A	Yes

Troubleshooting

Common Issues

MCP server connection failed

ktme mcp status
ktme mcp start

Service mapping not found

ktme mapping list
ktme mapping add my-service --file ~/path/to/docs/README.md

Enable debug logging

export KTME_LOG_LEVEL="debug"
ktme --verbose generate --commit abc123 --service my-service

Implementation Analysis & Roadmap

Codebase Architecture

ktme follows a well-structured architecture with clear separation of concerns:

• src/main.rs: Entry point with CLI command definitions
• src/cli/: Command-line interface and command handlers
• src/mcp/: MCP server implementation with tools
• src/git/: Git operations and diff extraction
• src/ai/: AI provider integrations (OpenAI, Claude)
• src/storage/: SQLite database models and repositories
• src/config/: Configuration management
• src/doc/: Documentation providers and writers
• src/error.rs: Comprehensive error handling

Current Build Status

• ✅ Project compiles successfully with cargo build --release
• ⚠️ 103 warnings (mostly unused code from incomplete features)
• ✅ All dependencies properly configured
• ✅ Binary runs and responds to commands
• ✅ Git operations work correctly
• ✅ MCP server starts in STDIO mode

Implementation Priorities

Phase 1: Complete Core Features (1-2 weeks)

• PR Extraction Implementation

  • Add GitHub API client for PR diffs
  • Implement GitLab API integration
  • Test with real pull requests

• Enhance Documentation Generation

  • Improve AI prompt templates
  • Add better diff-to-documentation logic
  • Support multiple documentation formats

• Complete MCP Tools

  • Enhance generate_documentation with AI integration
  • Add error handling for all tools
  • Test with Claude Code/Cursor integration

Phase 2: Provider Integration (2-3 weeks)

• Confluence Provider

  • Implement REST API client
  • Add authentication (token/oauth)
  • Create and update pages

• Template System

  • Implement variable substitution
  • Add built-in templates
  • Support custom templates

• Configuration Commands

  • Complete config command implementations
  • Add interactive setup
  • Validate configurations

Phase 3: Advanced Features (4-6 weeks)

• Knowledge Search & RAG

  • SQLite document caching
  • FTS5 full-text search
  • Vector embeddings for semantic search

• Feature Mapping System

  • Feature-to-service mapping
  • Feature-to-document linking
  • Team-based organization

• Confluence Sync

  • Incremental sync from Confluence
  • Bi-directional updates
  • Conflict resolution

Testing Strategy

The project currently lacks unit tests. Recommended testing approach:

• Unit Tests: Test individual components (Git reader, AI client, storage)
• Integration Tests: Test CLI commands end-to-end
• MCP Tests: Test MCP server with actual AI assistants
• E2E Tests: Full workflow from Git diff to documentation

Development Guidelines

• Keep It Simple: Avoid over-engineering, use straightforward solutions
• Incremental Development: Build and test one feature at a time
• Error Handling: Provide clear error messages for debugging
• Documentation: Add inline docs for complex logic
• Performance: Profile before optimizing, focus on actual bottlenecks

Contributing

Contributions are welcome!

• Fork the repository
• Create a feature branch (git checkout -b feature/amazing-feature)
• Commit your changes (git commit -m 'Add amazing feature')
• Push to the branch (git push origin feature/amazing-feature)
• Open a Pull Request

Development

# Build
cargo build

# Build in release mode (recommended for testing)
cargo build --release

# Run tests
cargo test

# Format code
cargo fmt

# Lint
cargo clippy

# Quick development cycle
cargo build --release && ./target/release/ktme --version

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgments

• Model Context Protocol for the MCP specification
• Rust MCP SDK for MCP server implementation
• git2-rs for Git integration
• clap for CLI parsing

Built with Rust. Powered by AI. Documentation made simple.