Haystack Docs MCP

[Hansehart]

Is your feature request related to a problem? Please describe.
Developing with Haystack often requires manual context-gathering — looking through documentation, examples, and configs to understand how to implement or connect components. This slows down iteration and increases the cognitive load when experimenting with pipelines or new features.

Describe the solution you'd like
I'd love to see an implementation similar to shadcn/ui MCP or Stripe MCP (ressources section). This would enable subagents that automatically retrieve relevant documentation, examples, and configuration details in real time when developers are building or debugging Haystack pipelines.

Describe alternatives you've considered
Manual context management for my subagents, fetch website, link local source code

Additional context
This would help developers focus on logic rather than lookup, making Haystack more accessible, extensible, and developer-friendly.

[julian-risch]

Hi @Hansehart , Thank you for your suggestion! We're currently working on further improving our documentation. While I can't give you a timeline yet, making Haystack documentation accessible via MCP is something we would like to do, yes.
As a first step, we're making all documentation files part of the Haystack GitHub repository. For example, Agent documentation now lives here: https://github.com/deepset-ai/haystack/blob/main/docs-website/versioned_docs/version-2.18/concepts/agents.mdx

[Hansehart]

Thank you for your feedback. This would be realy helpful, because then you could define a subagent e. g. in claude code like this:

---
name: Haystack Developer
description: You are responsible for integrating Haystack
tools: haystack-mcp  # <- What this Issue #9885 aims for
model: sonnet  # Optional - specify model alias or 'inherit'
---

When invoked:
- Elaborate the issue
- Identify the Haystack components or patterns needed
- Pull working examples and configurations
- Build production-ready implementation

Checklist:
- Code is simple and readable
- Functions and variables are well-named
- No duplicated code
- Proper error handling
- No exposed secrets or API keys
- Input validation implemented
- Good test coverage
- Performance considerations addressed

etc.

See more: Claude Code Subagents

I understand an MCP Server would require significant work to implement. However, I believe in specialized subagents and think this could help Haystack grow its adoption. When I started with Haystack, it was hard to get started at first and took some iterations to achieve satisfying results. An MCP would lower the barrier by providing on-demand working examples tailored to specific needs, complementing the existing cookbooks (which would still serve as sources for the MCP).

I would love to see Haystack grow. Happy to help test or provide feedback if you move forward with this!