RAGFlow MCP

[jbeckton]

Is the RAGFlow MCP server supposed to fetch the dataset ids internally? There is a method in the MCP server implementation but it returns an empty collection so the LLM cannot use the MCP server unless I dig up the dataset id and provide them in the prompt.

Seems like a big mess, is anyone using this in production?

[dosubot[bot]]

This is a confirmed bug in v0.24.0 — the list_datasets() method was passing None values for optional parameters to the /datasets API endpoint, causing server-side validation failures and returning an empty collection [1] [2].

It was fixed in PR #13123 (merged February 12, 2026) by conditionally adding parameters only when they have values [3].

To resolve this, upgrade to a newer build (nightly or any release after Feb 12, 2026) that includes the fix [4].

For context, dataset IDs are designed to be optional — when not provided, the MCP server should automatically fetch and search across all available datasets for the authenticated user [5] [6]. This was intentionally made optional back in August 2025 because smaller LLMs would sometimes misformat dataset IDs when using the MCP server in tools like n8n [6]. So you shouldn't need to dig up and provide dataset IDs manually.

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}
   

[jbeckton]

I think this project should be labeled as Beta to be honest.