Add a cloudflare docs MCP server using autorag

[cmsparks]

Waiting on my autorag to finish indexing, but this should be good for a review.

This autorag is intentionally unauthenticated, because it's not linked to any cloudflare specific resources.

[irvinebroque]

lgtm, i wonder in practice if mcp clients end up using the optional arguments (especially scoreThreshold) erroneously, when they would have been better off omitting? but we'll learn from this

does make me wonder about instrumentation

[aninibread]

lgtm, i wonder in practice if mcp clients end up using the optional arguments (especially scoreThreshold) erroneously, when they would have been better off omitting? but we'll learn from this

does make me wonder about instrumentation

Thank you for putting this together!!

I also wonder how this would work in practice. With AI search the expectation is for the developer to preset that value from experimentation. Here are values I'd recommend to start out and people can tune:

• max_results: 10
• match_threshold: 0
  This ensures that something will definitely be retrieved, although might not be the most accurate.

[aninibread]

Suggested change

	This tool returns a number of results from a vector database. These are embedded as resources in the response and are plaintext doucments in a variety of formats.
	This tool returns a number of results from a vector database. These are embedded as resources in the response and are plaintext documents in a variety of formats.

[cmsparks]

lgtm, i wonder in practice if mcp clients end up using the optional arguments (especially scoreThreshold) erroneously, when they would have been better off omitting? but we'll learn from this
does make me wonder about instrumentation

Thank you for putting this together!!

I also wonder how this would work in practice. With AI search the expectation is for the developer to preset that value from experimentation. Here are values I'd recommend to start out and people can tune:

* max_results: 10

* match_threshold: 0
  This ensures that something will definitely be retrieved, although might not be the most accurate.

Yeah, I think I'm also a bit fuzzy about what the match_threshold looks like in practice, so also not sure if the LLM would use that parameter correctly. In my testing, it seems to not be doing anything super weird with that match_threshold parameter though and I get reasonably relevant results.