@promptlab/mcp

PromptLab MCP Server

A Model Context Protocol (MCP) server for PromptLab experiments, providing programmatic access to experiment management, model configuration, and prompt engineering tools.

Installation

npx @promptlab/mcp --secret your_promptlab_token

Usage

The MCP server requires a PromptLab authentication token. You can obtain this from your PromptLab account settings.

npx @promptlab/mcp --secret pl_your_token_here

Features

• Experiment Management: Create, update, delete, and retrieve experiments
• Model Configuration: Manage available models and experiment-specific model assignments
• Message & Block Management: Full CRUD operations for experiment messages and blocks
• Variant Management: Create and manage prompt variants for A/B testing
• Judge Configuration: Set up LLM, code, or human judges for experiment evaluation
• Provider Key Management: Manage API keys for different AI providers

MCP Tools

Experiments

• listExperiments - Get all experiments
• createExperiment - Create new experiment
• getExperimentDetails - Get experiment details
• updateExperiment - Update experiment properties
• deleteExperiment - Delete experiment

Models

• getUserModels - Get available models
• getExperimentModels - Get models for experiment
• updateExperimentModels - Update experiment models

Messages & Blocks

• getMessage - Get message with blocks and variants
• addBlock - Add block to message
• getBlock - Get block details
• updateBlock - Update block properties
• deleteBlock - Delete block

Variants

• addVariant - Add variant to block
• updateVariant - Update variant
• deleteVariant - Delete variant

Judges

• updateExperimentJudgeType - Set judge type
• updateExperimentLLMJudge - Configure LLM judge
• updateExperimentCodeJudge - Configure code judge

Provider Keys

• addProviderKey - Add API key
• removeProviderKey - Remove API key
• validateProviderKey - Validate API key

Requirements

• Node.js 18.0.0 or higher
• Valid PromptLab authentication token

License

This software is proprietary and confidential. Unauthorized copying, distribution, or use is strictly prohibited.

Support

For support and documentation, visit promptlab.ai