MCP Development Context Engine

An MCP server that tracks coding assistant activity across sessions and repositories. Connects directly to Claude Code (or any MCP client) via stdio transport.

What it does

• Tracks prompts sent to any coding assistant
• Tracks assistant responses
• Tracks file changes per repository (with diffs)
• Extracts tasks from responses (TODO, next steps, etc.)
• Shares context across assistants via MCP tools
• Stores everything locally in SQLite — no cloud, no telemetry

Setup (one time)

# 1. Install dependencies
cd mcp_server
pip3 install -r requirements.txt

# 2. Add to Claude Code (globally, once)
claude mcp add --scope user --transport stdio mcp-context -- python3 /absolute/path/to/mcp_server/server.py

That's it. Every future Claude Code session will have the tracking tools available.

Available tools

Tool	Description
start_tracking_session(assistant_name, repo_path)	Start tracking a session
end_tracking_session(session_id)	End a tracking session
log_user_prompt(session_id, content)	Log a user prompt
log_assistant_response(session_id, content)	Log an assistant response (auto-extracts tasks)
log_file_modification(repo_path, file_path, diff, assistant_name)	Log a file change
get_context(repo_path)	Get full context: last assistant, prompts, responses, diffs, tasks
get_tasks(repo_path)	Get task list for a repo

Verification

1. Open Claude Code
2. Run /mcp to verify tools are listed
3. Ask Claude Code to call start_tracking_session or get_context

Configuration

Variable	Default	Description
MCP_DB_DIR	./storage	Directory for SQLite database

Requirements

• Python 3.10+
• No OS-specific dependencies — works on macOS, Linux, and Windows