MCP-over-HTTP endpoints removed from the API gateway - Cherry Studio

What changed

The v1 API server exposed MCP (Model Context Protocol) over HTTP for agents. The v2 API gateway (ported in #15705) does not carry these endpoints over, so they now return 404:

GET /v1/mcps — list active MCP servers
GET /v1/mcps/:server_id — MCP server info + tool list
ALL /v1/claw/:agentId/claw-mcp — the Claw agent's Streamable-HTTP MCP transport (session create/init/timeout/close + JSON-RPC dispatch)

A v1 agent-deletion hook for tearing down a Claw MCP server (cleanupClawServer) is also absent — though it was already uncalled dead code in the v1 tree, so its removal orphans no live call site.

Why this matters to the user

Any external client or agent that talked to Cherry Studio's MCP servers over HTTP — listing MCP servers, fetching their tools, or connecting to the Claw agent's MCP transport — will get a 404 against the v2 gateway. In-app MCP usage (MCP servers consumed inside Cherry Studio's own AI runtime) is unaffected; only the HTTP-exposed surface for external agents was removed.

What the user should do

TBD — pending product decision on whether MCP-over-HTTP returns in a later v2 release. For now there is no HTTP replacement; MCP must be consumed in-app rather than over the gateway.

Notes for release manager

The v2 data services backing the read endpoints still exist (mcpServerService, McpCatalogService), so /v1/mcps and /v1/mcps/:id are a thin re-port if we decide to restore them. The Claw Streamable-HTTP transport is the heavier piece — it needs its session/lifecycle machinery rebuilt in the gateway. (The old cleanupClawServer deletion hook was already dead code, so there is no existing hook to re-home — a restored transport would need its teardown written fresh.) Confirm with the agents owner whether any shipped agent flow depends on the Claw HTTP transport before finalizing this as a permanent removal.