Hayhooks

Overview#

Hayhooks is the official tool from deepset for serving Haystack pipelines as production-ready APIs. It supports one-click deployment of pipelines and agents as REST APIs, with native MCP Protocol support for seamless integration with MCP clients like Cursor and Claude Desktop.

Key Features#

REST API Deployment#

• Deploy Haystack pipelines and agents as REST APIs
• Support both YAML-based and wrapper-based deployment
• Auto-generate OpenAI-compatible /chat/completions endpoints

MCP Protocol Support#

• Run as MCP Server, exposing each pipeline/agent as an MCP Tool
• Integrate with MCP clients like Cursor and Claude Desktop
• Control core APIs via conversation (deploy, undeploy, list, run pipelines)

Frontend Integration#

• Chainlit UI: Embedded chat frontend, zero-config, streaming output, custom components
• Open WebUI: OpenAI-compatible backend with event notifications and status feedback

Other Capabilities#

• Built-in file upload handling for RAG systems and document processing
• Multi-component streaming with async_streaming_generator and streaming_generator

Typical Use Cases#

1. RAG System Deployment - Quickly deploy retrieval-augmented generation pipelines as APIs
2. AI Agent Servitization - Wrap Haystack agents as callable REST/MCP tools
3. Multi-LLM Streaming Pipelines - Chain multiple LLM components with automatic streaming
4. Document Indexing & Query - Use Elasticsearch for document indexing and semantic search

Installation & Quick Start#

# Basic installation
pip install hayhooks

# MCP support
pip install "hayhooks[mcp]"

# Chainlit UI
pip install "hayhooks[chainlit]"

# Start server
hayhooks run

# With Chainlit UI
hayhooks run --with-chainlit

Core Classes & API#

• BasePipelineWrapper - Base class for pipeline wrappers
• setup() - Initialize pipeline/agent
• run_api() / run_api_async() - Custom HTTP POST endpoints
• run_chat_completion_async() - OpenAI-compatible endpoints
• async_streaming_generator() - Async streaming generator
• streaming_generator() - Sync streaming generator

HTTP API Examples#

# Custom endpoint
curl -X POST http://localhost:1416/my_agent/run \
  -H 'Content-Type: application/json' \
  -d '{"question": "What can you do?"}'

# OpenAI-compatible endpoint
curl -X POST http://localhost:1416/chat/completions \
  -H 'Content-Type: application/json' \
  -d '{"model": "my_agent", "messages": [{"role": "user", "content": "Hello"}]}'

Project Info#

• Version: v1.13.0 (Beta)
• License: Apache-2.0
• Python Version: >=3.10, <3.15
• Deployment: Docker Compose, direct Python execution