refactor(mcp): flatten to mcp_server package, drop src layout

Rename the import package surfsense_mcp -> mcp_server and remove the
src/ layer so the project mirrors the backend's shape (project folder
!= package name, e.g. surfsense_backend/app). Kills the redundant
surfsense_mcp/src/surfsense_mcp nesting.

Distribution name and console command (surfsense-mcp) are unchanged;
only python -m and internal import paths move to mcp_server. Dockerfile
CMD updated; no PYTHONPATH added since the editable install already
makes the package importable.

surfsense_mcp/mcp_server/features/workspaces/__init__.py

@@ -0,0 +1,100 @@
+"""Search-space selector: discover workspaces and choose the active one.
+
+A workspace (the product calls it a "search space") scopes every other tool.
+These two tools let a client list what's available and pick one by name, so the
+rest of the conversation needs no ids.
+"""
+
+from __future__ import annotations
+
+from typing import Annotated
+
+from mcp.server.fastmcp import FastMCP
+from mcp.types import ToolAnnotations
+from pydantic import Field
+
+from ...core.rendering import ResponseFormatParam, to_json
+from ...core.workspace_context import Workspace, WorkspaceContext
+
+_READ_ONLY = ToolAnnotations(
+    readOnlyHint=True, destructiveHint=False, idempotentHint=True, openWorldHint=False
+)
+
+
+def register(mcp: FastMCP, context: WorkspaceContext) -> None:
+    """Register the workspace selector tools on the server."""
+
+    @mcp.tool(
+        name="surfsense_list_workspaces",
+        title="List workspaces",
+        annotations=_READ_ONLY,
+        structured_output=False,
+    )
+    async def list_workspaces(
+        response_format: ResponseFormatParam = "markdown",
+    ) -> str:
+        """List the SurfSense workspaces (search spaces) the account can access.
+
+        Use this to discover which workspaces exist before selecting one, or
+        when the user asks what search spaces they have. Returns each
+        workspace's name, id, description, ownership, and member count, and
+        marks the currently active one.
+        """
+        workspaces = await context.fetch_all()
+        if response_format == "json":
+            return to_json([_as_dict(w) for w in workspaces])
+        return _render_list(workspaces, active=context.active)
+
+    @mcp.tool(
+        name="surfsense_select_workspace",
+        title="Select active workspace",
+        annotations=_READ_ONLY,
+        structured_output=False,
+    )
+    async def select_workspace(
+        workspace: Annotated[
+            str,
+            Field(
+                description="Workspace name or numeric id; matching is "
+                "case-insensitive and a unique partial name works. "
+                "Example: 'Research'."
+            ),
+        ],
+    ) -> str:
+        """Set the active workspace (search space) that later tools default to.
+
+        Use this when the user says which search space to work in ("use my
+        Research space"), or after surfsense_list_workspaces when several
+        exist. Once set, workspace-scoped tools use it unless given a
+        different 'workspace'. Do NOT call it before every tool — once per
+        session is enough.
+        """
+        selected = await context.resolve(workspace)
+        return (
+            f"Active workspace is now '{selected.name}' (id {selected.id}). "
+            "Other tools will use it unless you pass a different 'workspace'."
+        )
+
+
+def _render_list(workspaces: list[Workspace], *, active: Workspace | None) -> str:
+    if not workspaces:
+        return "No accessible workspaces."
+    lines = ["# Workspaces", ""]
+    for workspace in workspaces:
+        marker = " — active" if active and active.id == workspace.id else ""
+        role = "owner" if workspace.is_owner else "member"
+        lines.append(f"- **{workspace.name}** (id {workspace.id}){marker}")
+        if workspace.description:
+            lines.append(f"  - {workspace.description}")
+        lines.append(f"  - {role}, {workspace.member_count} member(s)")
+    return "\n".join(lines)
+
+
+def _as_dict(workspace: Workspace) -> dict:
+    return {
+        "id": workspace.id,
+        "name": workspace.name,
+        "description": workspace.description,
+        "is_owner": workspace.is_owner,
+        "member_count": workspace.member_count,
+    }