+1 and thanks for the initative Tanishq ! This is a real good feature and long overdue!
I agree with denys , we should have seperate mcp for hs2 and hms. I have seem people run standalone hms without hs2. Maybe HS2 mcp can be superset of hms mcp? Regarding where the code resides, i don't have an opinion. Thanks! Simhadri G On Tue, Aug 25, 2026, 3:21 PM Shohei Okumiya <[email protected]> wrote: > Hi Tanishq, > > Thanks for driving the initiative. I'm happy to see HIVE-29370 and > HIVE-29681, which add new values to Hive Metastore. I think Hive > should become an Open Agentic Cloud-Native Lakehouse Platform; this > initiative will fill the last piece :) > > > 1. Should we have separate MCP servers for Metadata (HMS) & Query > Execution (HS2), or one MCP server that can serve both? > > I am biased toward a unified MCP, letting Hive operators and users > choose which features are enabled via configuration and OAuth 2 > scopes, for two reasons. First, as Tanishq says, SELECT queries can be > useful even for finding metadata. Second, although Lakehouse started > as query-engine agnostic, recent important topics such as table > maintenance, Server-Side planning, materialized views, etc. are > coupled with some compute engines by design. I guess we might be able > to provide an integrated experience if the MCP server can access both > HS2 and HMS. I'm not so confident, though. > > > 2. Where do we plan to store the MCP code? Should we store it in our > Apache/Hive repository itself or in a separate repository? > > I'd say starting in apache/hive is easy, though I don't have a strong > opinion. > > Best, > Okumin > > > On Tue, Aug 25, 2026 at 5:20 PM Denys Kuzmenko <[email protected]> > wrote: > > > > Hi Tanishq, > > > > +1 for the initiative and the overall direction. > > > > One point I would consider is keeping the HMS MCP separate from the Hive > query execution MCP. HMS is already a standalone component that can be > deployed and released independently of Hive, and it is used by engines > beyond Hive itself, such as Spark, Trino, and Impala. > > > > Because of that, an HMS-focused MCP could provide metadata discovery > capabilities as a reusable integration point for the broader ecosystem, > without coupling those capabilities to HiveServer2 or query execution. > > > > The two MCP servers could still share common code or APIs where > appropriate, but having independently deployable MCP endpoints seems more > aligned with the existing HMS architecture and its role as a shared > metadata service. > > > > Best, > > Denys >
