context-switcher-mcp

democratize-technology/context-switcher-mcp

3.2

If you are the rightful owner of context-switcher-mcp and would like to certify it and/or have it hosted online, please leave a comment on the right or send an email to henry@mcphub.com.

Context-Switcher MCP is a server that facilitates multi-perspective analysis through parallel thread orchestration, allowing for comprehensive problem evaluation from various viewpoints.

Tools
6
Resources
0
Prompts
0

Context-Switcher MCP

License: MIT Python 3.10+ MCP Compatible Tests

A Model Context Protocol (MCP) server that enables multi-perspective analysis through parallel thread orchestration. Break free from single-perspective thinking by analyzing problems through multiple lenses simultaneously.

🚀 Production Ready: Comprehensive test suite (1298+ tests), dependency vulnerabilities resolved, and battle-tested in real-world scenarios.

Overview

Context-Switcher MCP implements the thread orchestration pattern to provide parallel evaluation from different viewpoints. Each perspective can contribute insights or abstain with [NO_RESPONSE] when not relevant.

Features

  • Parallel Perspective Analysis: Evaluate problems through multiple viewpoints simultaneously
  • Flexible Abstention: Perspectives can opt out with [NO_RESPONSE] when not applicable
  • Multi-Provider Support: Works with AWS Bedrock, LiteLLM, and Ollama
  • Session Management: Maintain context across multiple analyses
  • Default Perspectives: Technical, Business, User Experience, and Risk
  • Custom Perspectives: Add domain-specific viewpoints on the fly
  • Synthesis: Find patterns and insights across all perspectives

Installation

Prerequisites

  • Python 3.10 or higher
  • MCP-compatible client (e.g., Claude Desktop, Cline)
  • At least one LLM provider configured:
    • AWS credentials for Bedrock
    • API keys for LiteLLM-supported models
    • Ollama running locally

Quick Start

# Install from PyPI (when published)
pip install context-switcher-mcp

# Or install from GitHub
pip install git+https://github.com/democratize-technology/context-switcher-mcp.git

Install from source

# Clone the repository
git clone https://github.com/democratize-technology/context-switcher-mcp.git
cd context-switcher-mcp

# Install the package
pip install -e .

Install for development

pip install -e ".[dev]"

Configuration

Claude Desktop

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "context-switcher": {
      "command": "python",
      "args": ["-m", "context_switcher_mcp"]
    }
  }
}

Cline

Add to your Cline MCP settings:

{
  "context-switcher": {
    "command": "python",
    "args": ["-m", "context_switcher_mcp"]
  }
}

Alternative Installation with UV

If you're using uv for Python package management, you can use the provided run script:

# Clone the repository
git clone https://github.com/democratize-technology/context-switcher-mcp.git
cd context-switcher-mcp

# Make the run script executable (if not already)
chmod +x run.sh

# The run script will handle everything
./run.sh

Claude Desktop Configuration with run.sh

For a more robust setup using the run script:

{
  "mcpServers": {
    "context-switcher": {
      "command": "/path/to/context-switcher-mcp/run.sh"
    }
  }
}

The run script will:

  • Create a virtual environment if needed
  • Install/update dependencies
  • Start the MCP server

Debug Mode

To enable debug logging:

DEBUG=1 ./run.sh

Or in Claude Desktop config:

{
  "mcpServers": {
    "context-switcher": {
      "command": "/path/to/context-switcher-mcp/run.sh",
      "env": {
        "DEBUG": "1"
      }
    }
  }
}

LLM Provider Setup

AWS Bedrock

Ensure AWS credentials are configured:

aws configure
LiteLLM

Set appropriate API keys:

export OPENAI_API_KEY="your-key"
# or
export ANTHROPIC_API_KEY="your-key"
Ollama

Ensure Ollama is running:

ollama serve
ollama pull llama3.2  # or your preferred model

Usage

Starting an Analysis

Use context-switcher to analyze database architecture choices

The tool will initialize with default perspectives: technical, business, user, and risk.

Adding Custom Perspectives

Add a "maintainability" perspective focusing on long-term code health

Running Analysis

How should we handle user authentication in our new microservice?

Each perspective will evaluate the question, providing insights or abstaining if not relevant.

Synthesizing Results

Synthesize the perspectives to find key insights

This will analyze patterns, conflicts, and unexpected connections across all viewpoints.

Available Tools

  1. start_context_analysis - Initialize multi-perspective analysis
  2. add_perspective - Add custom viewpoints dynamically
  3. analyze_from_perspectives - Broadcast questions to all perspectives
  4. synthesize_perspectives - Find patterns across viewpoints
  5. list_sessions - View all active analysis sessions
  6. get_session - Get details of a specific session

Default Perspectives

  • Technical: Architecture, implementation, scalability, performance
  • Business: ROI, market impact, strategic alignment, costs
  • User: Experience, usability, accessibility, workflow impact
  • Risk: Security, compliance, operational risks, privacy

Examples

Architecture Decision

Start context analysis for "choosing between REST and GraphQL"
Add "developer experience" perspective
Analyze: What are the tradeoffs between REST and GraphQL for our team?
Synthesize the results

Feature Evaluation

Start context analysis for "implementing real-time notifications"
Analyze: Should we build real-time notifications in-house or use a service?

Debugging Perspective Blindness

Start context analysis for "slow API response times"
Add "infrastructure" perspective
Add "database optimization" perspective
Analyze: Why are our API endpoints taking 5+ seconds to respond?

Development

Running Tests

pytest tests/

Code Quality

ruff check .
ruff format .

Thread Orchestration Pattern

This MCP implements the thread orchestration pattern with:

  • Parallel execution of all perspectives
  • NoResponse handling for selective participation
  • Provider flexibility per thread
  • Conversation history per perspective

Credits

Inspired by the thread orchestration pattern from multz and the thinking tools ecosystem including rubber-duck-mcp, sequential-thinking-mcp, and recursive-companion-mcp.

Security

This project follows security best practices:

  • Dependency Scanning: Regular security audits with pip-audit and safety
  • Input Validation: All user inputs are validated and sanitized
  • Secret Management: Secure handling of API keys and credentials
  • Vulnerability Reporting: See for reporting security issues

Reporting Vulnerabilities

Please report security vulnerabilities to code-developer@democratize.technology. See our for details.

Contributing

We welcome contributions! Please see our contribution guidelines:

  1. Fork the repository
  2. Create a feature branch (git checkout -b feature/amazing-feature)
  3. Make your changes with comprehensive tests
  4. Run the test suite: pytest tests/
  5. Ensure code quality: ruff check . && ruff format .
  6. Commit with descriptive messages
  7. Push to your fork and submit a Pull Request

Development Setup

# Clone and setup for development
git clone https://github.com/democratize-technology/context-switcher-mcp.git
cd context-switcher-mcp
pip install -e ".[dev]"
pre-commit install

Support

License

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

Acknowledgments

Inspired by the thread orchestration pattern from multz and the thinking tools ecosystem including rubber-duck-mcp, sequential-thinking-mcp, and recursive-companion-mcp.

Democratize Technology - Making advanced AI tools accessible to everyone.

Visit us at democratize.technology

Bedrock Model Configuration

The Context-Switcher MCP defaults to using us.anthropic.claude-3-7-sonnet-20250219-v1:0. To use a different model:

Option 1: Specify model when starting analysis

start_context_analysis(
    topic="Your topic",
    model_backend="bedrock",
    model_name="us.anthropic.claude-opus-4-20250514-v1:0"
)

Option 2: Environment variable (add to run.sh or Claude config)

export BEDROCK_MODEL_ID="us.anthropic.claude-opus-4-20250514-v1:0"
./run.sh

Available Bedrock Models

Anthropic Models (Inference Profiles):

  • us.anthropic.claude-3-7-sonnet-20250219-v1:0 (default)
  • us.anthropic.claude-opus-4-20250514-v1:0
  • us.anthropic.claude-sonnet-4-20250514-v1:0
  • us.anthropic.claude-3-5-haiku-20241022-v1:0

Meta Llama Models:

  • meta.llama3-3-70b-instruct-v1:0
  • meta.llama3-1-70b-instruct-v1:0
  • meta.llama3-1-8b-instruct-v1:0

Note: Use inference profile IDs (with "us." prefix) for Anthropic models to support on-demand throughput.

LiteLLM Configuration

To use your LiteLLM instance at 192.168.5.111:4000:

Get the API key from Kubernetes:

# Find the secret name
kubectl get secrets | grep litellm

# Extract the API key
kubectl get secret <secret-name> -o jsonpath='{.data.api-key}' | base64 -d

Configure the endpoint:

export LITELLM_API_BASE="http://192.168.5.111:4000"
export LITELLM_API_KEY="<your-api-key>"

Use in Context-Switcher:

start_context_analysis(
    topic="Your topic",
    model_backend="litellm",
    model_name="gpt-4"  # or any model available on your LiteLLM instance
)

Ollama Configuration

To use your Ollama instance at 192.168.1.250:

Configure the endpoint:

export OLLAMA_HOST="http://192.168.1.250:11434"  # Replace with actual port

Use in Context-Switcher:

start_context_analysis(
    topic="Your topic",
    model_backend="ollama",
    model_name="llama3.2"  # or any model available on your Ollama instance
)

Note: Ollama requires no authentication.