Python agent framework
Haystack
Open source AI orchestration framework for building production ready LLM applications with modular pipelines and agents in Python.
Quick decision
What it is
Haystack is an open source AI orchestration framework for building production ready LLM applications in Python, according to its GitHub repository. It is designed for modular pipelines and agent workflows with explicit control over retrieval, routing, memory, and generation.
The project positions itself around context engineering, with scalable RAG systems, multimodal applications, semantic search, question answering, and autonomous agents in a transparent architecture. Its README highlights agents built for production, native async support, and modular components for retrieval, indexing, tool calling, memory, and evaluation.
Haystack states it is model and vendor agnostic, integrating with providers such as OpenAI, Mistral, Anthropic, Cohere, Hugging Face, Google, Azure OpenAI, and AWS Bedrock. It also notes Hayhooks can serve Haystack pipelines and agents as REST APIs or MCP servers.
What it does
- Builds modular pipelines and agent workflows with explicit control over retrieval, routing, memory, and generation
- Supports RAG, multimodal applications, semantic search, and question answering use cases
- Provides built in components for retrieval, indexing, tool calling, memory, and evaluation, plus custom components
- Runs pipelines synchronously or asynchronously with streaming, and agents with concurrent tool calls
- Integrates with many model providers and infrastructure components without locking to one vendor
What it does not do
- Not a hosted model provider; it orchestrates models and components you configure
- Not a single purpose MCP server builder; MCP serving is handled through Hayhooks and integrations
- Not a TypeScript framework; the core project is Python based
Setup
Install Haystack and run an MCP toolset agent
Install from PyPI, then connect an MCP server through MCPToolset as shown in the Haystack MCP docs.
pip install haystack-ai mcp-haystack
from haystack_integrations.tools.mcp import MCPToolset, StdioServerInfo
toolset = MCPToolset(server_info=StdioServerInfo(command="uvx", args=["mcp-server-time"]))Environment variables (names only)
OPENAI_API_KEY
Source: https://docs.haystack.deepset.ai/docs/mcptoolset. Examples use placeholders only.
MCP support: Haystack documents MCPToolset, which dynamically discovers and loads tools from any MCP compliant server over Streamable HTTP, SSE (deprecated), or StdIO, and the README notes Hayhooks can expose pipelines and agents as MCP servers.
Health and maintenance
26,692
Checked 2026-10-07
3,253
Checked 2026-10-07
Stars, forks, Apache-2.0 license, and release count read from the GitHub repository page fetched 2026-10-07. PyPI release URL showed haystack-ai 3.3.0.
Verified compatible tools
Via tavily-haystack (TavilyWebSearch). Official Haystack integrations page says the Tavily integration provides TavilyWebSearch, searches the web using the Tavily API, returns Haystack Document objects, needs a Tavily API key, and installs with pip install tavily-haystack. The Haystack WebSearch docs also list TavilyWebSearch. Evidence. Setup detail
Via brave-search-haystack (BraveWebSearch). Official Haystack integrations page describes Brave Search with Haystack, provides BraveWebSearch, says it searches using the Brave Search API, needs a Brave Search API key, and installs with pip install brave-search-haystack. Haystack WebSearch docs list BraveWebSearch. Evidence. Setup detail
Via FirecrawlWebSearch. Haystack WebSearch documentation lists FirecrawlWebSearch as a search engine using the Firecrawl API, alongside TavilyWebSearch and BraveWebSearch. Evidence. Setup detail
Grouped view: tools for Haystack.
Sources and freshness
- GitHub repository: https://github.com/deepset-ai/haystack
- MCPToolset documentation: https://docs.haystack.deepset.ai/docs/mcptoolset
- WebSearch documentation: https://docs.haystack.deepset.ai/docs/websearch
- PyPI haystack-ai: https://pypi.org/pypi/haystack-ai/json
Last checked 2026-10-07. Verification label: source verified. It does not mean AgentsUse ran the framework. Spotted an error? Send a correction. Back to the framework index.