haystack MCP Client: Orchestrating LLM Context with Explicit Control
haystack MCP Client: Orchestrating LLM Context with Explicit Control
haystack is an open-source AI orchestration framework, implemented as an MCP Client in Python, designed for developers building production-ready LLM applications. It provides explicit control over the context engineering process, allowing for the creation of modular pipelines and agent workflows.
Granular Control Over Context Engineering
haystack emphasizes transparency and traceability in how information is handled before reaching the model. Developers can define pipelines and agent workflows with precise control over retrieval, ranking, filtering, combining, structuring, and routing of context. This modular approach ensures that every step—from memory access to tool invocation and generation—is explicit and manageable.
Production-Ready Agent Capabilities
The framework's agents are built for production environments, offering lifecycle hooks like before_llm, before_tool, and on_exit for implementing guardrails and custom logic. Monitoring and cost control are integrated, with out-of-the-box tracking for step_count, token_usage, and tool calls. haystack also supports progressive skill discovery via SkillToolset, ensuring skill descriptions are only introduced into context when relevant. Developers can leverage ready-made agents from Agent Pack, such as a deep research agent or an advanced RAG agent, or build their own.
Modular Design and Asynchronous Execution
haystack provides built-in components for common tasks like retrieval, indexing, tool calling, memory, and evaluation, while also allowing for custom component creation. Its design supports complex logic within pipelines and agent workflows, including loops, branches, and conditional execution. The framework features native async support, enabling a single Pipeline to run synchronously or asynchronously and stream tokens, and Agent to execute concurrent tool calls. This client is also model- and vendor-agnostic, providing flexibility in LLM integration.