Bedrock Prompts MCP Server

# Changelog All notable changes to the Bedrock Prompts MCP Server. ## [0.2.0] - 2024-11-14 ### Added - **Batch Invocation**: New `batch_invoke_bedrock_prompt` tool for parallel execution - Run same prompt with multiple variable sets - Configurable parallelism (1-10 workers) - Separate tracking of successes/failures - Ideal for generating multiple survey responses or testing variations - **Streaming Support**: New `invoke_bedrock_prompt_stream` tool - Real-time text streaming as model generates - Returns both chunks and full completion - Improved UX for long-form content - Works with Claude and Titan models - **Multi-Model Support**: Automatic detection and formatting for: - Amazon Titan (titan-text-express, titan-text-lite) - Meta Llama (llama-2, llama-3) - Mistral AI (mistral-7b, mixtral-8x7b) - Cohere (command, command-light) - AI21 Labs (jurassic-2) - Anthropic Claude (existing) - **Model-Specific Request Builders**: - `build_request_body_*()` functions for each model type - Automatic detection via `get_model_type()` - Correct parameter mapping per model API - **Model-Specific Response Parsers**: - `parse_response_*()` functions for each model type - Handles different response formats automatically - Extracts text from various response structures - **Documentation**: - `QUICK_REFERENCE.md`: Common use cases and best practices - `examples.py`: Runnable examples for all features - Enhanced README with batch/streaming examples ### Changed - `invoke_prompt()` now uses model-agnostic request builder - Better model ID extraction from Bedrock ARNs - More detailed response metadata including model type - Improved error messages with context ### Technical Details - Added ThreadPoolExecutor for parallel batch processing - Streaming uses `invoke_model_with_response_stream` API - Timeout handling (120s per invocation in batch mode) - Better exception handling across all model types ## [0.1.0] - 2024-11-13 ### Added - Initial MCP server implementation - `list_bedrock_prompts`: Browse available prompts - `get_bedrock_prompt_details`: View prompt configuration - `invoke_bedrock_prompt`: Execute prompts with variables - `list_bedrock_prompt_versions`: Version management - Claude Desktop integration - Basic Claude model support - Variable substitution ({{var}} and {var} syntax) - AWS credential handling - Error handling and logging ### Technical Details - Python 3.10+ required - boto3 for AWS SDK - mcp library for protocol implementation - stdio-based communication