Teleprompter

An MCP server that manages and exposes tools to allow prompt re-use with LLMs.

Table of Contents

• Features

• MCP Configuration

• Usage Examples

• Environment Variables

• Testing

• Contributing

• License

• Acknowledgements

Features

• Prompt Storage & Reuse: Store, search, and retrieve prompt templates for LLMs.

• MCP Server: Exposes prompt tools via the Model Context Protocol (MCP).

• Prompt Variables: Supports template variables (e.g., {{name}}) for dynamic prompt generation.

• Search: Fast full-text search over stored prompts using MiniSearch.

• TypeScript: Modern, type-safe codebase.

• Extensive Testing: Includes unit and integration tests with Vitest.

MCP Configuration

To use Teleprompter with your LLM client, add this configuration:

Note: Replace /path/to/your/prompts-directory with the absolute path where you want prompts stored.

Usage Examples

Once configured, you can use Teleprompter with your LLM by using prompt tags in your conversations. Here's a detailed example that shows how it solves the problem of repeating complex instructions:

🎵 Music Discovery on Spotify

The Problem: Every time you want music recommendations, you have to remind your LLM of all your preferences and constraints:

• "Don't suggest songs I already have in my playlists"

• "Avoid explicit lyrics"

• "Add songs to my queue for review, not directly to playlists"

• "Focus on discovering new artists, not just popular hits"

• "Consider my current activity and mood"

• "Provide brief explanations for why each song fits"

The Solution: Create a prompt that captures all these instructions once.

Creating the prompt: Ask your LLM: "Create a prompt called 'spotify-discover' that helps me find new music with all my specific preferences and workflow requirements."

This creates a comprehensive template like:

Using it:

Now you just fill in your current mood and activity, and get perfectly tailored recommendations that follow all your rules—no need to repeat your constraints every time.

🔄 Other Common Use Cases

📋 Work Ticket Management

• Create prompts for JIRA/Linear ticket formatting with your team's specific requirements

• Include standard fields, priority levels, acceptance criteria templates

• Avoid repeating your company's ticket standards every time

📧 Email Templates

• Customer support responses with your company's tone and required disclaimers

• Follow-up sequences that match your communication style

• Automated inclusion of signatures, links, and standard information

📝 Code Review Guidelines

• Technical review checklists with your team's specific standards

• Security considerations and performance criteria

• Documentation requirements and testing expectations

The common thread: stop repeating yourself. If you find yourself giving the same detailed instructions to your LLM repeatedly, create a prompt for it.

🔍 Discovering Existing Prompts

You can search your prompt library:

Or list all available prompts:

✏️ Manual Editing

Prompts are stored as simple markdown files in your PROMPT_STORAGE_PATH directory. You can also create and edit them directly with your favorite text editor:

• Each prompt is saved as {id}.md in your prompts directory

• Use {{variable_name}} syntax for template variables

• Standard markdown formatting is supported

• File changes are automatically picked up by the server

💡 Best Practices

1. Use descriptive IDs: Choose prompt IDs that clearly indicate their purpose (e.g., meeting-notes, code-review-checklist)

2. Include helpful variables: Use {{variable_name}} for dynamic content that changes each time you use the prompt

3. Organize by category: Consider using prefixes like task-, content-, analysis- to group related prompts

Testing

Run all tests:

Run tests with coverage:

Tests are written with Vitest. Coverage reports are generated in the coverage/ directory.

Contributing

Contributions are welcome! Please:

• Follow the existing code style (see .prettierrc.json and .eslintrc.mjs).

• Add tests for new features or bug fixes.

License

This project is licensed under the MIT License. See LICENSE for details.

Acknowledgements

• Model Context Protocol (MCP)

• MiniSearch

• Vitest

• Zod

• dotenv

Made with ❤️ by John Anderson