mcp-instructions-server

A minimal, production-ready Model Context Protocol (MCP) server in Python that provides canonical instructions to any MCP-capable AI application.

Installation

Ensure you have Python 3.10+ installed. Then install the package:

pip install -e .

Or using pip directly:

pip install .

Running the Server

Development Mode

python -m mcp_instructions_server.server

Or using the dev script (after installation):

dev

Production Mode

start

Development Scripts

• dev - Run the server in development mode
• start - Start the server in production mode
• lint - Run linting checks using ruff
• format - Format code using ruff formatter

Example MCP Client Usage

Fetching Prompts

The server provides three prompts:

1. security - Security instruction for enforcing least privilege and security best practices
2. rai_guidance - Responsible AI (RAI) guidance for ethical AI practices
3. all_instructions - Combined security and RAI guidance instructions

Example using an MCP client:

# Example: Fetching the security prompt
# The client would call the MCP prompt API to get:
# {
#   "role": "system",
#   "content": "Enforce least privilege, redact secrets, refuse exploit/abuse content, validate untrusted inputs, prefer most restrictive rule on conflict."
# }

Fetching Resources

The server exposes one resource:

• instructions://canonical - Complete canonical instructions in Markdown format

Example:

# The client would call the MCP resource API to fetch:
# URI: instructions://canonical
# MIME Type: text/markdown
# Content: Markdown with version header and both Security and RAI Guidance sections

Health Check Tool

The server provides a health_check tool that returns server status:

# Example response:
{
  "status": "ok",
  "serverTime": "2024-01-01T12:00:00.000000Z"
}

Example Cursor Configuration

To integrate the mcp-instructions-server with Cursor, add the following to your Cursor configuration:

{
  "mcpServers": {
    "instructions": {
      "command": "python",
      "args": ["-m", "mcp_instructions_server.server"]
    }
  }
}

Note: Adjust the path to Python as needed. If you've installed the package in a virtual environment, you may need to use the full path to the Python executable.

Acceptance Checks

Prompts Panel

Verify that the prompts panel displays:

• security
• rai_guidance
• all_instructions

Resource Fetch

Ensure that fetching the instructions://canonical resource returns:

• Markdown content with both Security and RAI Guidance sections
• Version header: # Instructions Version v1.0.0

Health Check

Confirm that calling health_check returns:

• {"status": "ok", "serverTime": "<ISO8601>"}

Project Structure

mcp_instructions_server/
├── __init__.py          # Package initialization
├── server.py            # Main entrypoint; registers prompts, resources, tools
├── instructions.py      # Stores versioned instruction text
├── prompts.py           # Wires MCP prompts
├── resources.py         # Wires MCP resources
└── tools.py             # Defines health_check tool

Quality Assurance

• Code Quality: Clean, well-documented code with no TODOs
• Type Hints: Full type annotations throughout
• Logging: Structured logging with graceful shutdown handling
• Linting: Run lint to check code quality
• Formatting: Run format to format code

Updating Instructions

To update the instruction content, edit mcp_instructions_server/instructions.py:

• Update INSTRUCTIONS_VERSION as needed
• Modify SECURITY_INSTRUCTION for security guidelines
• Modify RAI_GUIDANCE for RAI guidance
• The ALL_INSTRUCTIONS_MD will automatically include both sections

License

MIT