diff --git a/internal/index/index.go b/internal/index/index.go
index 67e0845..10475aa 100644
--- a/internal/index/index.go
+++ b/internal/index/index.go
@@ -3,6 +3,7 @@ package index
 import (
 	"fmt"
 	"os"
+	"path/filepath"
 	"strings"
 
 	"github.com/marte-dev/marte-dev-tools/internal/parser"
@@ -13,6 +14,26 @@ type ProjectTree struct {
 	References []Reference
 }
 
+func (pt *ProjectTree) ScanDirectory(rootPath string) error {
+	return filepath.Walk(rootPath, func(path string, info os.FileInfo, err error) error {
+		if err != nil {
+			return err
+		}
+		if !info.IsDir() && strings.HasSuffix(info.Name(), ".marte") {
+			content, err := os.ReadFile(path)
+			if err != nil {
+				return err // Or log and continue
+			}
+			p := parser.NewParser(string(content))
+			config, err := p.Parse()
+			if err == nil {
+				pt.AddFile(path, config)
+			}
+		}
+		return nil
+	})
+}
+
 type Reference struct {
 	Name     string
 	Position parser.Position
diff --git a/internal/lsp/server.go b/internal/lsp/server.go
index 4e9813e..5f311e6 100644
--- a/internal/lsp/server.go
+++ b/internal/lsp/server.go
@@ -10,6 +10,7 @@ import (
 
 	"github.com/marte-dev/marte-dev-tools/internal/index"
 	"github.com/marte-dev/marte-dev-tools/internal/parser"
+	"github.com/marte-dev/marte-dev-tools/internal/validator"
 )
 
 type JsonRpcMessage struct {
@@ -26,6 +27,11 @@ type JsonRpcError struct {
 	Message string `json:"message"`
 }
 
+type InitializeParams struct {
+	RootURI  string `json:"rootUri"`
+	RootPath string `json:"rootPath"`
+}
+
 type DidOpenTextDocumentParams struct {
 	TextDocument TextDocumentItem `json:"textDocument"`
 }
@@ -60,6 +66,31 @@ type TextDocumentIdentifier struct {
 	URI string `json:"uri"`
 }
 
+type DefinitionParams struct {
+	TextDocument TextDocumentIdentifier `json:"textDocument"`
+	Position     Position               `json:"position"`
+}
+
+type ReferenceParams struct {
+	TextDocument TextDocumentIdentifier `json:"textDocument"`
+	Position     Position               `json:"position"`
+	Context      ReferenceContext       `json:"context"`
+}
+
+type ReferenceContext struct {
+	IncludeDeclaration bool `json:"includeDeclaration"`
+}
+
+type Location struct {
+	URI   string `json:"uri"`
+	Range Range  `json:"range"`
+}
+
+type Range struct {
+	Start Position `json:"start"`
+	End   Position `json:"end"`
+}
+
 type Position struct {
 	Line      int `json:"line"`
 	Character int `json:"character"`
@@ -74,6 +105,18 @@ type MarkupContent struct {
 	Value string `json:"value"`
 }
 
+type PublishDiagnosticsParams struct {
+	URI         string          `json:"uri"`
+	Diagnostics []LSPDiagnostic `json:"diagnostics"`
+}
+
+type LSPDiagnostic struct {
+	Range    Range  `json:"range"`
+	Severity int    `json:"severity"`
+	Message  string `json:"message"`
+	Source   string `json:"source"`
+}
+
 var tree = index.NewProjectTree()
 
 func RunServer() {
@@ -121,6 +164,22 @@ func readMessage(reader *bufio.Reader) (*JsonRpcMessage, error) {
 func handleMessage(msg *JsonRpcMessage) {
 	switch msg.Method {
 	case "initialize":
+		var params InitializeParams
+		if err := json.Unmarshal(msg.Params, &params); err == nil {
+			root := ""
+			if params.RootURI != "" {
+				root = uriToPath(params.RootURI)
+			} else if params.RootPath != "" {
+				root = params.RootPath
+			}
+			
+			if root != "" {
+				fmt.Fprintf(os.Stderr, "Scanning workspace: %s\n", root)
+				tree.ScanDirectory(root)
+				tree.ResolveReferences()
+			}
+		}
+
 		respond(msg.ID, map[string]any{
 			"capabilities": map[string]any{
 				"textDocumentSync":   1, // Full sync
@@ -130,7 +189,7 @@ func handleMessage(msg *JsonRpcMessage) {
 			},
 		})
 	case "initialized":
-		// Do nothing
+		runValidation("")
 	case "shutdown":
 		respond(msg.ID, nil)
 	case "exit":
@@ -160,6 +219,16 @@ func handleMessage(msg *JsonRpcMessage) {
 			fmt.Fprint(os.Stderr, "not recovered hover parameters\n")
 			respond(msg.ID, nil)
 		}
+	case "textDocument/definition":
+		var params DefinitionParams
+		if err := json.Unmarshal(msg.Params, &params); err == nil {
+			respond(msg.ID, handleDefinition(params))
+		}
+	case "textDocument/references":
+		var params ReferenceParams
+		if err := json.Unmarshal(msg.Params, &params); err == nil {
+			respond(msg.ID, handleReferences(params))
+		}
 	}
 }
 
@@ -174,6 +243,7 @@ func handleDidOpen(params DidOpenTextDocumentParams) {
 	if err == nil {
 		tree.AddFile(path, config)
 		tree.ResolveReferences()
+		runValidation(params.TextDocument.URI)
 	}
 }
 
@@ -188,9 +258,78 @@ func handleDidChange(params DidChangeTextDocumentParams) {
 	if err == nil {
 		tree.AddFile(path, config)
 		tree.ResolveReferences()
+		runValidation(params.TextDocument.URI)
 	}
 }
 
+func runValidation(uri string) {
+	v := validator.NewValidator(tree)
+	v.ValidateProject()
+	v.CheckUnused()
+
+	// Group diagnostics by file
+	fileDiags := make(map[string][]LSPDiagnostic)
+	
+	// Collect all known files to ensure we clear diagnostics for fixed files
+	knownFiles := make(map[string]bool)
+	collectFiles(tree.Root, knownFiles)
+	
+	// Initialize all known files with empty diagnostics
+	for f := range knownFiles {
+		fileDiags[f] = []LSPDiagnostic{}
+	}
+
+	for _, d := range v.Diagnostics {
+		severity := 1 // Error
+		if d.Level == validator.LevelWarning {
+			severity = 2 // Warning
+		}
+
+		diag := LSPDiagnostic{
+			Range: Range{
+				Start: Position{Line: d.Position.Line - 1, Character: d.Position.Column - 1},
+				End:   Position{Line: d.Position.Line - 1, Character: d.Position.Column - 1 + 10}, // Arbitrary length
+			},
+			Severity: severity,
+			Message:  d.Message,
+			Source:   "mdt",
+		}
+		
+		path := d.File
+		if path != "" {
+			fileDiags[path] = append(fileDiags[path], diag)
+		}
+	}
+
+	// Send diagnostics for all known files
+	for path, diags := range fileDiags {
+		fileURI := "file://" + path
+		notification := JsonRpcMessage{
+			Jsonrpc: "2.0",
+			Method:  "textDocument/publishDiagnostics",
+			Params:  mustMarshal(PublishDiagnosticsParams{
+				URI:         fileURI,
+				Diagnostics: diags,
+			}),
+		}
+		send(notification)
+	}
+}
+
+func collectFiles(node *index.ProjectNode, files map[string]bool) {
+	for _, frag := range node.Fragments {
+		files[frag.File] = true
+	}
+	for _, child := range node.Children {
+		collectFiles(child, files)
+	}
+}
+
+func mustMarshal(v any) json.RawMessage {
+	b, _ := json.Marshal(v)
+	return b
+}
+
 func handleHover(params HoverParams) *Hover {
 	path := uriToPath(params.TextDocument.URI)
 	line := params.Position.Line + 1
@@ -239,6 +378,93 @@ func handleHover(params HoverParams) *Hover {
 	}
 }
 
+func handleDefinition(params DefinitionParams) any {
+	path := uriToPath(params.TextDocument.URI)
+	line := params.Position.Line + 1
+	col := params.Position.Character + 1
+
+	res := tree.Query(path, line, col)
+	if res == nil {
+		return nil
+	}
+
+	var targetNode *index.ProjectNode
+	if res.Reference != nil && res.Reference.Target != nil {
+		targetNode = res.Reference.Target
+	} else if res.Node != nil {
+		targetNode = res.Node
+	}
+
+	if targetNode != nil {
+		var locations []Location
+		for _, frag := range targetNode.Fragments {
+			if frag.IsObject {
+				locations = append(locations, Location{
+					URI: "file://" + frag.File,
+					Range: Range{
+						Start: Position{Line: frag.ObjectPos.Line - 1, Character: frag.ObjectPos.Column - 1},
+						End:   Position{Line: frag.ObjectPos.Line - 1, Character: frag.ObjectPos.Column - 1 + len(targetNode.RealName)},
+					},
+				})
+			}
+		}
+		return locations
+	}
+
+	return nil
+}
+
+func handleReferences(params ReferenceParams) []Location {
+	path := uriToPath(params.TextDocument.URI)
+	line := params.Position.Line + 1
+	col := params.Position.Character + 1
+
+	res := tree.Query(path, line, col)
+	if res == nil {
+		return nil
+	}
+
+	var targetNode *index.ProjectNode
+	if res.Node != nil {
+		targetNode = res.Node
+	} else if res.Reference != nil && res.Reference.Target != nil {
+		targetNode = res.Reference.Target
+	}
+
+	if targetNode == nil {
+		return nil
+	}
+
+	var locations []Location
+	if params.Context.IncludeDeclaration {
+		for _, frag := range targetNode.Fragments {
+			if frag.IsObject {
+				locations = append(locations, Location{
+					URI: "file://" + frag.File,
+					Range: Range{
+						Start: Position{Line: frag.ObjectPos.Line - 1, Character: frag.ObjectPos.Column - 1},
+						End:   Position{Line: frag.ObjectPos.Line - 1, Character: frag.ObjectPos.Column - 1 + len(targetNode.RealName)},
+					},
+				})
+			}
+		}
+	}
+
+	for _, ref := range tree.References {
+		if ref.Target == targetNode {
+			locations = append(locations, Location{
+				URI: "file://" + ref.File,
+				Range: Range{
+					Start: Position{Line: ref.Position.Line - 1, Character: ref.Position.Column - 1},
+					End:   Position{Line: ref.Position.Line - 1, Character: ref.Position.Column - 1 + len(ref.Name)},
+				},
+			})
+		}
+	}
+
+	return locations
+}
+
 func formatNodeInfo(node *index.ProjectNode) string {
 	class := node.Metadata["Class"]
 	if class == "" {
@@ -261,8 +487,8 @@ func formatNodeInfo(node *index.ProjectNode) string {
 		}
 
 		// Size
-		dims := node.Metadata["NumberOfDimensions"]
-		elems := node.Metadata["NumberOfElements"]
+	dims := node.Metadata["NumberOfDimensions"]
+elems := node.Metadata["NumberOfElements"]
 		if dims != "" || elems != "" {
 			sigInfo += fmt.Sprintf("**Size**: `[%s]`, `%s` dims ", elems, dims)
 		}
@@ -287,4 +513,4 @@ func respond(id any, result any) {
 func send(msg any) {
 	body, _ := json.Marshal(msg)
 	fmt.Printf("Content-Length: %d\r\n\r\n%s", len(body), body)
-}
+}
\ No newline at end of file
diff --git a/internal/parser/parser.go b/internal/parser/parser.go
index cc29d3b..2a11217 100644
--- a/internal/parser/parser.go
+++ b/internal/parser/parser.go
@@ -8,8 +8,7 @@ import (
 
 type Parser struct {
 	lexer    *Lexer
-	tok      Token
-	peeked   bool
+	buf      []Token
 	comments []Comment
 	pragmas  []Pragma
 }
@@ -21,21 +20,23 @@ func NewParser(input string) *Parser {
 }
 
 func (p *Parser) next() Token {
-	if p.peeked {
-		p.peeked = false
-		return p.tok
+	if len(p.buf) > 0 {
+		t := p.buf[0]
+		p.buf = p.buf[1:]
+		return t
 	}
-	p.tok = p.fetchToken()
-	return p.tok
+	return p.fetchToken()
 }
 
 func (p *Parser) peek() Token {
-	if p.peeked {
-		return p.tok
+	return p.peekN(0)
+}
+
+func (p *Parser) peekN(n int) Token {
+	for len(p.buf) <= n {
+		p.buf = append(p.buf, p.fetchToken())
 	}
-	p.tok = p.fetchToken()
-	p.peeked = true
-	return p.tok
+	return p.buf[n]
 }
 
 func (p *Parser) fetchToken() Token {
@@ -85,11 +86,30 @@ func (p *Parser) parseDefinition() (Definition, error) {
 	tok := p.next()
 	switch tok.Type {
 	case TokenIdentifier:
-		// field = value
+		// Could be Field = Value OR Node = { ... }
 		name := tok.Value
 		if p.next().Type != TokenEqual {
-			return nil, fmt.Errorf("%d:%d: expected =", p.tok.Position.Line, p.tok.Position.Column)
+			return nil, fmt.Errorf("%d:%d: expected =", tok.Position.Line, tok.Position.Column)
 		}
+
+		// Disambiguate based on RHS
+		nextTok := p.peek()
+		if nextTok.Type == TokenLBrace {
+			// Check if it looks like a Subnode (contains definitions) or Array (contains values)
+			if p.isSubnodeLookahead() {
+				sub, err := p.parseSubnode()
+				if err != nil {
+					return nil, err
+				}
+				return &ObjectNode{
+					Position: tok.Position,
+					Name:     name,
+					Subnode:  sub,
+				}, nil
+			}
+		}
+
+		// Default to Field
 		val, err := p.parseValue()
 		if err != nil {
 			return nil, err
@@ -99,11 +119,12 @@ func (p *Parser) parseDefinition() (Definition, error) {
 			Name:     name,
 			Value:    val,
 		}, nil
+
 	case TokenObjectIdentifier:
 		// node = subnode
 		name := tok.Value
 		if p.next().Type != TokenEqual {
-			return nil, fmt.Errorf("%d:%d: expected =", p.tok.Position.Line, p.tok.Position.Column)
+			return nil, fmt.Errorf("%d:%d: expected =", tok.Position.Line, tok.Position.Column)
 		}
 		sub, err := p.parseSubnode()
 		if err != nil {
@@ -119,6 +140,42 @@ func (p *Parser) parseDefinition() (Definition, error) {
 	}
 }
 
+func (p *Parser) isSubnodeLookahead() bool {
+	// We are before '{'.
+	// Look inside:
+	// peek(0) is '{'
+	// peek(1) is first token inside
+	
+	t1 := p.peekN(1)
+	if t1.Type == TokenRBrace {
+		// {} -> Empty. Assume Array (Value) by default, unless forced? 
+		// If we return false, it parses as ArrayValue.
+		// If user writes "Sig = {}", is it an empty signal?
+		// Empty array is more common for value. 
+		// If "Sig" is a node, it should probably have content or use +Sig.
+		return false 
+	}
+	
+	if t1.Type == TokenIdentifier {
+		// Identifier inside.
+		// If followed by '=', it's a definition -> Subnode.
+		t2 := p.peekN(2)
+		if t2.Type == TokenEqual {
+			return true
+		}
+		// Identifier alone or followed by something else -> Reference/Value -> Array
+		return false
+	}
+	
+	if t1.Type == TokenObjectIdentifier {
+		// +Node = ... -> Definition -> Subnode
+		return true
+	}
+	
+	// Literals -> Array
+	return false
+}
+
 func (p *Parser) parseSubnode() (Subnode, error) {
 	tok := p.next()
 	if tok.Type != TokenLBrace {
diff --git a/internal/validator/validator.go b/internal/validator/validator.go
index b723908..c686314 100644
--- a/internal/validator/validator.go
+++ b/internal/validator/validator.go
@@ -59,7 +59,6 @@ func (v *Validator) validateNode(node *index.ProjectNode) {
 	}
 
 	// Check for mandatory Class if it's an object node (+/$)
-	// Root node usually doesn't have a name or is implicit
 	if node.RealName != "" && (node.RealName[0] == '+' || node.RealName[0] == '$') {
 		hasClass := false
 		hasType := false
@@ -102,12 +101,75 @@ func (v *Validator) validateNode(node *index.ProjectNode) {
 	}
 }
 
-// Legacy/Compatibility method if needed, but we prefer ValidateProject
-func (v *Validator) Validate(file string, config *parser.Configuration) {
-	// No-op or local checks if any
+func (v *Validator) CheckUnused() {
+	referencedNodes := make(map[*index.ProjectNode]bool)
+	for _, ref := range v.Tree.References {
+		if ref.Target != nil {
+			referencedNodes[ref.Target] = true
+		}
+	}
+
+	v.checkUnusedRecursive(v.Tree.Root, referencedNodes)
 }
 
-func (v *Validator) CheckUnused() {
-	// To implement unused check, we'd need reference tracking in Index
-	// For now, focusing on duplicate fields and class validation
+func (v *Validator) checkUnusedRecursive(node *index.ProjectNode, referenced map[*index.ProjectNode]bool) {
+	// Heuristic for GAM
+	if isGAM(node) {
+		if !referenced[node] {
+			v.Diagnostics = append(v.Diagnostics, Diagnostic{
+				Level:    LevelWarning,
+				Message:  fmt.Sprintf("Unused GAM: %s is defined but not referenced in any thread or scheduler", node.RealName),
+				Position: v.getNodePosition(node),
+				File:     v.getNodeFile(node),
+			})
+		}
+	}
+
+	// Heuristic for DataSource and its signals
+	if isDataSource(node) {
+		for _, signal := range node.Children {
+			if !referenced[signal] {
+				v.Diagnostics = append(v.Diagnostics, Diagnostic{
+					Level:    LevelWarning,
+					Message:  fmt.Sprintf("Unused Signal: %s is defined in DataSource %s but never referenced", signal.RealName, node.RealName),
+					Position: v.getNodePosition(signal),
+					File:     v.getNodeFile(signal),
+				})
+			}
+		}
+	}
+
+	for _, child := range node.Children {
+		v.checkUnusedRecursive(child, referenced)
+	}
+}
+
+func isGAM(node *index.ProjectNode) bool {
+	if node.RealName == "" || (node.RealName[0] != '+' && node.RealName[0] != '$') {
+		return false
+	}
+	_, hasInput := node.Children["InputSignals"]
+	_, hasOutput := node.Children["OutputSignals"]
+	return hasInput || hasOutput
+}
+
+func isDataSource(node *index.ProjectNode) bool {
+	if node.Parent != nil && node.Parent.Name == "Data" {
+		return true
+	}
+	return false
+}
+
+func (v *Validator) getNodePosition(node *index.ProjectNode) parser.Position {
+	if len(node.Fragments) > 0 {
+		return node.Fragments[0].ObjectPos
+	}
+	return parser.Position{Line: 1, Column: 1}
+}
+
+func (v *Validator) getNodeFile(node *index.ProjectNode) string {
+	if len(node.Fragments) > 0 {
+		return node.Fragments[0].File
+	}
+	return ""
 }
\ No newline at end of file
diff --git a/mdt b/mdt
index 5a8d32b..f13bcd1 100755
Binary files a/mdt and b/mdt differ
diff --git a/specification.md b/specification.md
index 41599ad..ace1e8c 100644
--- a/specification.md
+++ b/specification.md
@@ -40,7 +40,8 @@ The LSP server should provide the following capabilities:
   - **URI Symbols**: The symbols `+` and `$` used for object nodes are **not** written in the URI of the `#package` macro (e.g., use `PROJECT.NODE` even if the node is defined as `+NODE`).
 - **Build Process**:
   - The build tool merges all files sharing the same base namespace.
-  - **Multi-File Nodes**: Nodes can be defined across multiple files. The build tool and validator must merge these definitions before processing.
+  - **Multi-File Definitions**: Nodes and objects can be defined across multiple files. The build tool, validator, and LSP must merge these definitions (including all fields and sub-nodes) from the entire project to create a unified view before processing or validating.
+  - **Global References**: References to nodes, signals, or objects can point to definitions located in any file within the project.
   - **Merging Order**: For objects defined across multiple files, the **first file** to be considered is the one containing the `Class` field definition.
   - **Field Order**: Within a single file, the relative order of defined fields must be maintained.
   - The LSP indexes only files belonging to the same project/namespace scope.
@@ -75,7 +76,7 @@ The LSP server should provide the following capabilities:
 ### Semantics
 
 - **Nodes (`+` / `$`)**: The prefixes `+` and `$` indicate that the node represents an object.
-  - **Constraint**: These nodes _must_ contain a field named `Class` within their subnode definition.
+  - **Constraint**: These nodes _must_ contain a field named `Class` within their subnode definition (across all files where the node is defined).
 - **Signals**: Signals are considered nodes but **not** objects. They do not require a `Class` field.
 - **Pragmas (`//!`)**: Used to suppress specific diagnostics. The developer can use these to explain why a rule is being ignored.
 - **Structure**: A configuration is composed by one or more definitions.
@@ -134,6 +135,9 @@ The tool must build an index of the configuration to support LSP features and va
 ### Validation Rules
 
 - **Consistency**: The `lsp`, `check`, and `build` commands **must share the same validation engine** to ensure consistent results across all tools.
+- **Global Validation Context**:
+  - All validation steps must operate on the aggregated view of the project.
+  - A node's validity is determined by the combination of all its fields and sub-nodes defined across all project files.
 - **Class Validation**:
   - For each known `Class`, the validator checks:
     - **Mandatory Fields**: Verification that all required fields are present.
@@ -144,7 +148,7 @@ The tool must build an index of the configuration to support LSP features and va
     - Class validation rules must be defined in a separate schema file.
     - **Project-Specific Classes**: Developers can define their own project-specific classes and corresponding validation rules, expanding the validation capabilities for their specific needs.
 - **Duplicate Fields**:
-  - **Constraint**: A field must not be defined more than once within the same object/node scope.
+  - **Constraint**: A field must not be defined more than once within the same object/node scope, even if those definitions are spread across different files.
   - **Multi-File Consideration**: Validation must account for nodes being defined across multiple files (merged) when checking for duplicates.
 
 ### Formatting Rules
diff --git a/test/integration/fmt.marte b/test/integration/fmt.marte
index ad28797..69b3cb7 100644
--- a/test/integration/fmt.marte
+++ b/test/integration/fmt.marte
@@ -1,7 +1,7 @@
 #package TEST.FMT
 
 // Detached comment
-
+//# Test
 +Node = {
     Class = "MyClass"