context-engine-mcp 2.2.1

MCP-based contextual flag system for AI assistants

Context Engine MCP

Note: This project was inspired by the pioneering work in SuperClaude Framework and SuperGemini Framework. Special thanks to SuperClaude-Org team members @NomenAK and @mithun50 whose work made this possible.

Context Engine provides 18 contextual flags that guide assistant behavior (e.g., --strict, --auto). It exposes an MCP stdio server and small setup helpers for common clients.

Quick Start

# Install (recommended)
pipx install context-engine-mcp

# For Claude Code
context-engine install

# For Continue 
context-engine install --target cn

Then in your client/assistant, use prompts with flags:

• "Fix this bug --auto" (auto-select flags)
• "--save" (handoff documentation)
• "Analyze --strict" (precise, zero-tolerance mode)

18 Flags

Flag	Purpose
--analyze	Multi-angle systematic analysis
--auto	AI selects optimal flag combination
--collab	Co-develop solutions through trust-based iteration
--concise	Minimal communication
--discover	Discover existing solutions before building new
--explain	Progressive disclosure
--git	Version control best practices
--lean	Essential focus only
--load	Load handoff documentation
--parallel	Multi-agent processing
--performance	Speed and efficiency optimization
--readonly	Analysis only mode
--refactor	Code quality improvement
--reset	Reset all flag states to new (clears session cache)
--save	Handoff documentation
--seq	Sequential thinking
--strict	Zero-error enforcement
--todo	Task management

Installation

Claude Code

# Install package
pipx install context-engine-mcp

# Install configuration files
context-engine install

Register the MCP server with Claude CLI:

# Choose ONE of these commands:

# Standard Python installation
claude mcp add -s user -- context-engine context-engine-mcp

# UV installation  
claude mcp add -s user -- context-engine uv run context-engine-mcp

# Custom command
claude mcp add -s user -- context-engine <your-command>

This also writes ~/.claude/CONTEXT-ENGINE.md and appends a reference to ~/.claude/CLAUDE.md.

Continue Extension

# Install package
pipx install context-engine-mcp

# Install configuration files
context-engine install --target cn

Edit ~/.continue/mcpServers/context-engine.yaml and uncomment ONE option:

# Option 1: Standard Python (most common)
name: Context Engine MCP
command: context-engine-mcp

# Option 2: UV installation  
# name: Context Engine MCP
# command: uv
# args: ["run", "context-engine-mcp"]

# Option 3: Custom installation
# name: Context Engine MCP  
# command: <your-custom-command>

Restart VS Code, then type @ in Continue chat to access MCP tools.

Gemini CLI

# Install package
pipx install context-engine-mcp

# Install configuration files for Gemini CLI
context-engine install --target gemini-cli

This command:

• Appends @CONTEXT-ENGINE.md to ~/.gemini/GEMINI.md (adds once; no duplicate)
• Writes latest instructions to ~/.gemini/CONTEXT-ENGINE.md

It does not modify ~/.gemini/settings.json. If required, register the MCP stdio command in Gemini CLI settings:

• Command: context-engine-mcp
• Args: []
• Transport: stdio

MCP registration (example)

• File: ~/.gemini/settings.json
• Add or merge this into the mcpServers section:

{
  "mcpServers": {
    "context-engine": {
      "type": "stdio",
      "command": "context-engine-mcp",
      "args": [],
      "env": {}
    }
  }
}

Gemini CLI settings

• Location: ~/.gemini/settings.json
• Structure:

{
  "mcpServers": {
    "<server-name>": {
      "type": "stdio",
      "command": "<executable or interpreter>",
      "args": ["<arg1>", "<arg2>", "..."],
      "env": { "ENV_KEY": "value" }
    }
  }
}

Common setups

• pipx (PATH):

{
  "mcpServers": {
    "context-engine": {
      "type": "stdio",
      "command": "context-engine-mcp",
      "args": [],
      "env": {}
    }
  }
}

• uv (run):

{
  "mcpServers": {
    "context-engine": {
      "type": "stdio",
      "command": "uv",
      "args": ["run", "context-engine-mcp"],
      "env": {}
    }
  }
}

• venv absolute path (Windows):

{
  "mcpServers": {
    "context-engine": {
      "type": "stdio",
      "command": "C:\\path\\to\\venv\\Scripts\\context-engine.exe",
      "args": [],
      "env": {}
    }
  }
}

• venv with interpreter (module run):

{
  "mcpServers": {
    "context-engine": {
      "type": "stdio",
      "command": "/path/to/venv/bin/python",
      "args": ["-m", "context_engine_mcp"],
      "env": {}
    }
  }
}

Notes

• type is stdio.
• If the command is not on PATH, use an absolute path (escape backslashes on Windows).
• After editing, restart Gemini CLI and verify tools (e.g., list_available_flags).

Usage

In Chat

# Auto mode - AI selects flags
"Refactor this code --auto"

# Direct flags
"--save"  # Creates handoff doc
"--analyze --strict"  # Multi-angle analysis with zero errors
"--reset --analyze"  # Reset session and reapply

# Combined flags
"Review this --analyze --strict --seq"

MCP Tools

• list_available_flags() - Shows all 18 flags
• get_directives(['--flag1', '--flag2']) - Activates flags

Development: use pip install -e . for editable installs.

Configuration updates: edit ~/.context-engine/flags.yaml and restart the MCP server.

Optional MCP Servers

Additional MCP servers can complement certain flags:

For --seq flag:

# Sequential thinking server  
claude mcp add -s user -- sequential-thinking npx -y @modelcontextprotocol/server-sequential-thinking

These are optional; Context Engine works without them.

Session

• Duplicate flags produce a brief reminder instead of repeating full directives.
• Use --reset when the task/context changes (resets all flag states to new).
• The server tracks active flags per session.
• Note: In Claude, flag states persist through /clear or /compact commands. Use --reset to reinitialize.

--auto

--auto instructs the assistant to analyze the task and pick appropriate flags (do not include --auto in get_directives calls).

Behavior

• --auto only: the assistant selects a full set of flags automatically.
• --auto --flag1 --flag2: the assistant applies --flag1, --flag2 and may add additional flags if helpful. User‑specified flags take priority when there is overlap or conflict.
• --flag1 --flag2 (without --auto): only the specified flags are applied.

Files Created

~/.claude/
├── CLAUDE.md                     # References @CONTEXT-ENGINE.md
├── CONTEXT-ENGINE.md             # Flag instructions (auto-updated)
├── hooks/
│   └── context-engine-hook.py    # Hook for flag detection (Claude Code only)
└── settings.json                 # Updated with hook registration (Claude Code only)

~/.continue/
├── config.yaml         # Contains Context Engine rules
└── mcpServers/
    ├── context-engine.yaml
    ├── sequential-thinking.yaml
    └── context7.yaml

~/.context-engine/
└── flags.yaml          # Flag definitions

~/.gemini/
├── GEMINI.md           # References @CONTEXT-ENGINE.md
└── CONTEXT-ENGINE.md   # Flag instructions (auto-updated)

Uninstallation

# Complete uninstall from all environments (Claude Code + Continue)
context-engine uninstall

# Remove Python package
pipx uninstall context-engine-mcp

Note: During uninstallation, ~/.context-engine/flags.yaml is backed up to ~/flags.yaml.backup_YYYYMMDD_HHMMSS before removal. During installation, existing flags.yaml is backed up and updated to the latest version.

Claude Code note: Uninstall removes the @CONTEXT-ENGINE.md reference from ~/.claude/CLAUDE.md, deletes ~/.claude/CONTEXT-ENGINE.md if present, removes the hook file from ~/.claude/hooks/context-engine-hook.py, and removes the hook registration from ~/.claude/settings.json.

Gemini CLI note: Uninstall removes the @CONTEXT-ENGINE.md reference from ~/.gemini/GEMINI.md and deletes ~/.gemini/CONTEXT-ENGINE.md if present.

Continue note: Uninstall removes the Context Engine rules from ~/.continue/config.yaml (when present) and deletes ~/.continue/mcpServers/context-engine.yaml if present.

License

MIT