feat(mcp): add scheduling client contracts

Assisted-by: Hephaestus:openai/gpt-5.5
Signed-off-by: Owen Adirah <owenadira@gmail.com>

Author: owezzy

pkg/mcp/localaitools/client.go

@@ -64,6 +64,10 @@ type LocalAIClient interface {
 	// ---- System ----
 	SystemInfo(ctx context.Context) (*SystemInfo, error)
 	ListNodes(ctx context.Context) ([]Node, error)
+	ListScheduling(ctx context.Context) ([]ModelSchedulingConfig, error)
+	GetScheduling(ctx context.Context, modelName string) (*ModelSchedulingConfig, error)
+	SetScheduling(ctx context.Context, req SetSchedulingRequest) (*ModelSchedulingConfig, error)
+	DeleteScheduling(ctx context.Context, modelName string) error
 	// SetNodeVRAMBudget sets (or, with an empty budget, clears) a federated
 	// node's VRAM allocation cap as a sticky admin override. Only meaningful
 	// in distributed mode; single-process clients report it as unavailable.

pkg/mcp/localaitools/dto.go

@@ -1,6 +1,9 @@
 package localaitools
 
-import "github.com/mudler/LocalAI/core/services/voiceprofile"
+import (
+	"github.com/mudler/LocalAI/core/services/nodes"
+	"github.com/mudler/LocalAI/core/services/voiceprofile"
+)
 
 // DTOs for the LocalAIClient interface. Where the same shape already exists
 // elsewhere (config.Gallery, gallery.Metadata, schema.KnownBackend,
@@ -115,6 +118,31 @@ type SetNodeVRAMBudgetRequest struct {
 	Budget string `json:"budget,omitempty" jsonschema:"VRAM allocation cap as a percentage (e.g. 80%) or absolute amount (e.g. 12GB). Empty string clears the override."`
 }
 
+// ModelSchedulingConfig is the REST/MCP wire shape for one per-model
+// distributed scheduling rule. The NodeSelector field is a JSON object encoded
+// as a string, matching /api/nodes/scheduling responses.
+type ModelSchedulingConfig = nodes.ModelSchedulingConfig
+
+// SetSchedulingRequest is the input for set_scheduling. It mirrors
+// /api/nodes/scheduling so standalone MCP and REST callers preserve the same
+// PATCH-style semantics for the optional prefix-cache routing fields.
+type SetSchedulingRequest struct {
+	ModelName           string            `json:"model_name"                         jsonschema:"Installed model name whose distributed scheduling rule should be created or updated."`
+	NodeSelector        map[string]string `json:"node_selector,omitempty"            jsonschema:"Optional node-label selector. Empty means any healthy backend node."`
+	MinReplicas         int               `json:"min_replicas"                       jsonschema:"Minimum desired replicas. Mutually exclusive with spread_all."`
+	MaxReplicas         int               `json:"max_replicas"                       jsonschema:"Maximum desired replicas. Must be >= min_replicas when non-zero. Mutually exclusive with spread_all."`
+	SpreadAll           bool              `json:"spread_all,omitempty"               jsonschema:"When true, keep one replica on every matching node. Mutually exclusive with min_replicas/max_replicas."`
+	RoutePolicy         *string           `json:"route_policy,omitempty"             jsonschema:"Optional prefix-cache route policy override. Omit to preserve the existing value on updates."`
+	BalanceAbsThreshold *int              `json:"balance_abs_threshold,omitempty"    jsonschema:"Optional absolute imbalance threshold override. Omit to preserve the existing value on updates."`
+	BalanceRelThreshold *float64          `json:"balance_rel_threshold,omitempty"    jsonschema:"Optional relative imbalance threshold override. Omit to preserve the existing value on updates."`
+	MinPrefixMatch      *float64          `json:"min_prefix_match,omitempty"         jsonschema:"Optional minimum prefix match threshold override. Omit to preserve the existing value on updates."`
+}
+
+// DeleteSchedulingRequest identifies the model scheduling rule to remove.
+type DeleteSchedulingRequest struct {
+	ModelName string `json:"model_name" jsonschema:"Installed model name whose scheduling config should be removed."`
+}
+
 // ImportModelURIRequest is the input for import_model_uri. It mirrors the
 // REST surface (`/models/import-uri`) closely so both clients can produce
 // identical responses; the BackendPreference is a flat field rather than the

pkg/mcp/localaitools/httpapi/routes.go

@@ -24,6 +24,7 @@ const (
 	routeBackendsKnown   = "/backends/known"
 	routeBackendsApply   = "/backends/apply"
 	routeNodes           = "/api/nodes"
+	routeScheduling      = "/api/nodes/scheduling"
 	routeVRAMEstimate    = "/api/models/vram-estimate"
 	routeBranding        = "/api/branding"
 	routeSettings        = "/api/settings"
@@ -66,3 +67,7 @@ func routeVoiceProfileDelete(id string) string {
 func routeNodeVRAMBudget(id string) string {
 	return "/api/nodes/" + url.PathEscape(id) + "/vram-budget"
 }
+
+func routeModelScheduling(modelName string) string {
+	return "/api/nodes/scheduling/" + url.PathEscape(modelName)
+}

pkg/mcp/localaitools/tools.go

@@ -17,6 +17,8 @@ const (
 	ToolListKnownBackends   = "list_known_backends"
 	ToolSystemInfo          = "system_info"
 	ToolListNodes           = "list_nodes"
+	ToolListScheduling      = "list_scheduling"
+	ToolGetScheduling       = "get_scheduling"
 	ToolVRAMEstimate        = "vram_estimate"
 	ToolGetBranding         = "get_branding"
 	ToolGetUsageStats       = "get_usage_stats"
@@ -42,6 +44,8 @@ const (
 	ToolCreateVoiceProfile = "create_voice_profile"
 	ToolDeleteVoiceProfile = "delete_voice_profile"
 	ToolSetNodeVRAMBudget  = "set_node_vram_budget"
+	ToolSetScheduling      = "set_scheduling"
+	ToolDeleteScheduling   = "delete_scheduling"
 
 	// ToolListAliases is read-only but lives here so the alias tools stay
 	// grouped; the catalog tests assert its read-only placement.