godotlens-mcp
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| GODOT_LSP_HOST | No | Godot LSP server host | 127.0.0.1 |
| GODOT_LSP_PORT | No | Godot LSP server port | 6005 |
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
This server publishes no instructions, or was last inspected before Glama recorded them.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| gdscript_statusA | Check connection status to the Godot LSP server. Returns: connection status, host, and port. Use this to verify Godot editor is running before using other tools. If disconnected, start Godot editor with your project open. |
| gdscript_definitionA | Navigate to the definition of a symbol at a given position. Returns: file path and line number where the symbol is defined. IMPORTANT: Uses ZERO-BASED coordinates (editor line 1 = pass line 0). Use when you need to find where a function, variable, or class is defined. |
| gdscript_referencesA | Find all references to a symbol across the entire project. Returns: list of locations (file, line, character) where the symbol is used. IMPORTANT: Uses ZERO-BASED coordinates. Essential for impact analysis before refactoring: 'What code uses this symbol?' |
| gdscript_hoverA | Get type information and documentation for a symbol at a given position. Returns: type signature, documentation string, or description of the symbol. IMPORTANT: Uses ZERO-BASED coordinates. Use to understand what type a variable is, or what a function returns. |
| gdscript_symbolsA | List all symbols (classes, functions, variables, signals, enums) in a file. Returns: symbol tree with name, kind, and line number for each symbol. Use to understand the structure of a file before making changes. WORKFLOW: gdscript_symbols to explore, then gdscript_hover or gdscript_definition for details. |
| gdscript_signature_helpA | Get function signature and parameter information at a call site. Returns: function name, parameters with types, and return type. IMPORTANT: Uses ZERO-BASED coordinates. Use when you need to know the correct parameters for a function call. |
| gdscript_renameA | Rename a symbol across all files in the project. Returns: workspace edit with all changes needed. IMPORTANT: Uses ZERO-BASED coordinates. WORKFLOW: (1) gdscript_references to preview impact, (2) gdscript_rename to rename, (3) gdscript_sync_files to refresh LSP state. |
| gdscript_sync_fileA | Notify Godot's LSP that a file was modified and get updated diagnostics. Returns: diagnostics (errors/warnings) for the synced file. WHEN TO CALL: After using Edit/Write tools to modify a .gd file. The LSP does not watch files, so you must call this to refresh analysis. Optionally pass content directly to avoid reading from disk. |
| gdscript_sync_filesA | Batch sync multiple modified files with Godot's LSP. Returns: diagnostics for all synced files. WHEN TO CALL: After modifying multiple .gd files with Edit/Write tools. More efficient than calling gdscript_sync_file repeatedly. Reads content from disk for all specified files. |
| gdscript_release_fileA | Release a file from the LSP session, so Godot stops serving the copy this session opened and reads from disk again. Returns: whether the file was open, and what was actually done. WHEN TO CALL: after deleting a .gd file, or when you want the LSP to forget content you synced earlier. NOTE: Godot removed its file-deletion notification in 4.6, so this cannot purge project-wide state; that clears when the editor rescans. |
| gdscript_symbols_batchA | Get symbols from multiple files in a single call. Returns: map of file path to symbol tree for each file. More efficient than calling gdscript_symbols repeatedly. Use to understand the structure of multiple files at once. |
| gdscript_definitions_batchA | Get definitions for multiple symbol positions in a single call. Returns: list of definition locations for each position. IMPORTANT: Uses ZERO-BASED coordinates. More efficient than calling gdscript_definition repeatedly. |
| gdscript_references_batchA | Find references for multiple symbols in a single call. Returns: list of reference locations for each position. IMPORTANT: Uses ZERO-BASED coordinates. More efficient than calling gdscript_references repeatedly. Use for bulk impact analysis across multiple symbols. |
| gdscript_diagnosticsA | Get compiler errors and warnings for one or more files. Returns: list of diagnostics with line, severity (1=Error, 2=Warning, 3=Info, 4=Hint), and message. WORKFLOW: (1) Edit files, (2) gdscript_sync_files to refresh, (3) gdscript_diagnostics to check for errors. Use before committing to catch issues early. |
| gdscript_engine_apiA | Get authoritative documentation for a Godot ENGINE class or member, from the exact editor build in use. Returns: signature with argument names, types and defaults, plus documentation. USE THIS instead of recalling Godot's API from memory - it is the ground truth for the user's version and prevents inventing methods that do not exist. Pass 'member' for a specific method/property/signal; omit it to check a class exists. |
| gdscript_completeA | Get valid completions at a cursor position, from Godot's own completion engine. Returns: candidate labels with kind and detail. IMPORTANT: Uses ZERO-BASED coordinates. This is the only SCENE-AWARE query available: Godot resolves the scene that owns this script and completes against the real node, so $NodePath entries and the signals actually present on that node are included. No analysis of the .gd file alone can reproduce that. |
| gdscript_validateA | Check proposed file content for errors WITHOUT writing it to disk. Returns: valid flag, errors and warnings. WHEN TO CALL: before writing an edit, so broken code never reaches the project. The LSP is restored to the on-disk content afterwards. |
| gdscript_references_in_fileA | Find occurrences of a symbol within ONE file. Returns: list of line/char positions. IMPORTANT: Uses ZERO-BASED coordinates. Much cheaper than gdscript_references, which reparses every .gd file in the project on Godot 4.6+. Requires Godot 4.7+; reports unsupported otherwise. |
| scene_stateA | Get Godot's own resolved view of a scene: node tree with types, script attachments, unique_name_in_owner flags, exported property values, and the signal connections declared in the scene. Godot's LSP reads .gd files only, so NONE of this is visible to gdscript_references or gdscript_rename. Runs the engine to resolve the scene, so inherited scenes and instanced children are resolved the way Godot actually instantiates them. Requires a Godot binary (GODOT_BIN or ./godot/). |
| scene_validateA | Check a scene for what Godot gets silently wrong when it loads or runs it. Errors: connections to handlers or nodes that do not exist, signals the source does not have, handlers taking the wrong number of arguments, unknown node types, scripts whose base type does not fit the node, missing parents, duplicate names and duplicate unique names, and changes to nodes that do not exist in an instanced scene. Warnings: unknown properties, values of the wrong type, and NodePaths to nodes not in the scene. Returns: per-scene problems with kind, severity and detail; valid means no errors. WHEN TO CALL: after editing a .tscn, or after renaming or removing a signal handler in GDScript. Godot accepts all of these with no error or only a console warning. Facts come from the engine; handler existence from the LSP's parse of the attached script. |
| debug_statusA | Check the connection to Godot's debug adapter and report whether the game is running, paused, or finished. The adapter is served by the Godot editor on port 6006 and needs no addon. Use this first if any debug_* tool behaves unexpectedly. |
| debug_outputA | Read what a game started with debug_run produced: print() output and stdout/stderr lines, plus structured errors - push_error, push_warning, engine errors and runtime script errors - each with file, function, ZERO-BASED line and call stack. Returns: lines, errors, and their counts. THIS IS THE ONLY WAY to see what the game actually did; the language server reports whether code compiles, not what it printed. Output and errors are drained on each call, so successive calls return only what is new. |
| debug_set_breakpointsA | Set breakpoints in a GDScript file, replacing any previously set in that file. Returns: each breakpoint with whether Godot verified it and the line it bound to. IMPORTANT: lines are ZERO-BASED, matching every other tool here. Set these before running the game, then use debug_stack_trace and debug_inspect once execution stops. |
| debug_stack_traceA | Get the call stack where execution is currently paused. Returns: frames with function name, file and ZERO-BASED line, plus why it stopped. An empty frame list means execution is not paused - frames exist only while stopped at a breakpoint or a runtime error. |
| debug_inspectA | Inspect variables visible in a stack frame. Returns: each scope (locals, members, globals) with its variables, values and types. Use the frame_id from debug_stack_trace. Only meaningful while execution is paused. |
| debug_evaluateA | Evaluate a GDScript expression in the context of a paused frame. Returns: the resulting value and its type. Use to check state at a breakpoint without adding print() calls and re-running. |
| debug_continueB | Resume a paused game. Returns: confirmation. |
| debug_pauseA | Pause the running game. Returns: where it stopped. Use to inspect state at an arbitrary moment rather than a preset breakpoint. |
| debug_step_overA | Step over one line in the paused game. Returns: the new stop location. |
| debug_terminateB | Stop the running game. Returns: confirmation. |
| gdscript_findA | Find where a symbol is declared BY NAME, without needing its position. Returns: declaration sites with file, ZERO-BASED line and character, kind, and containing class. USE THIS FIRST when you know a name but not its location - the returned line/character feed directly into gdscript_references, gdscript_hover and gdscript_rename. Guessing a character offset and landing one column off returns an empty result that looks identical to 'no such symbol'. Positions come from the language server, not from text matching. |
| project_configA | Get the project's resolved configuration: autoload singletons, input action names, class_name globals, and the main scene. Autoload and input action names are BARE STRINGS at the point of use - GameState.add_score(1), Input.is_action_pressed("jump") - and nothing validates them. Neither the compiler nor the language server catches a typo; it is a silent runtime no-op. Check names here before writing them. Values come from ProjectSettings via the engine, so defaults and feature-tagged overrides resolve correctly. Requires a Godot binary. |
| project_validateA | Find references that no longer resolve, project-wide: scene and resource dependencies pointing at moved or deleted files, missing autoload scripts, and a missing main scene. Returns: broken references with the file that holds each, plus stale UIDs (Godot falls back to the text path and warns). WHEN TO CALL: after moving, renaming or deleting files. Nothing else reports these until the reference is loaded at runtime. Uses the engine's own resolution: a reference whose UID still resolves is not broken. Script preload()/extends paths are compile errors - gdscript_diagnostics reports them. Requires a Godot binary; runs no project code. |
| debug_runA | Run the project from the open Godot editor and collect what it prints. Returns: captured output, structured errors with call stacks, whether the game exited, and any stop reason. THIS CLOSES THE LOOP: edit, sync, run, and read the actual behaviour, without the developer pressing F5 or pasting console output back. The editor has ONE run slot: a second debug_run stops the first game. Set breakpoints with debug_set_breakpoints BEFORE calling this if you want execution to pause. Long-running games keep going - use debug_output to keep reading and debug_terminate to stop. |
| runtime_runA | Launch the game as its own process, connected straight back to GodotLens - no editor run slot, so several agents can each run one at once. Returns: session_id, and the output and structured errors from the first seconds. Runs HEADLESS by default (no window, no sound); pass windowed=true only when you need runtime_screenshot - the window is placed off-screen and takes no focus. Script errors are reported, never stop the game. Then use runtime_tree, runtime_inspect and runtime_output to see what the game is doing. The game runs through GodotLens's helper (loaded for this run only, nothing is written to the project), which adds runtime_eval and runtime_input. |
| runtime_stopA | Stop a game started with runtime_run. Returns: exit code. |
| runtime_statusA | List the games GodotLens is running. Returns: each session with pid, scene, whether it is headless, running and connected, plus this server's pid. |
| runtime_outputA | Read what a running game printed and every structured error it reported since the last call - push_error, push_warning, engine and script errors, each with file, function, ZERO-BASED line and call stack. Drained on each call. |
| runtime_treeA | The live scene tree of a running game, including nodes created at runtime. Returns: nested nodes with name, type (script class_name or path when scripted), path, object_id and visibility. Use the path or object_id with runtime_inspect. |
| runtime_inspectB | Current property values of one node or object in a running game, as the engine reports them. Returns: class and each property's type and value. |
| runtime_setA | Change a property on a node in a running game, then read it back. Values are converted to the property's own type: pass [1, 2] or 'Vector2(1, 2)' for a Vector2, '#ff0000' for a Color, a res:// path for a resource. Returns: the value the engine now reports. |
| runtime_evalA | Evaluate an expression in a running game, while it runs - no breakpoint needed. Uses Godot's Expression: calls, operators, member access and constructors, no statements or assignment. Names available: the base node's members (default the current scene), autoloads, engine singletons such as Input, and class_name globals. Calls run for real and may change the game. Returns: the value, or the engine's error text. |
| runtime_inputA | Press an input action in a running game, as a player would: pressed for the given number of frames, then released, through the normal input path, so is_action_just_pressed and _unhandled_input see it. Actions come from the project's Input Map (project_config lists them). Returns: once released. |
| runtime_reloadA | Hot-reload changed scripts into a running game, keeping its state. Refuses when the language server reports errors in them: a script that fails to compile is disabled in the running game. Covers .gd scripts only - edited .tscn files and new or changed assets need a restart. |
| runtime_pauseB | Freeze or unfreeze a running game's scene tree. Returns: the new state. |
| runtime_time_scaleB | Speed up or slow down a running game. Returns: the new scale. |
| runtime_stepA | Advance a paused game by one frame. Windowed sessions only: a headless game never draws a frame, and Godot re-pauses after the next drawn frame. |
| runtime_monitorsA | Performance monitors of a running game, about once a second: FPS, frame times, object, node and orphan-node counts, memory, draw calls, physics, plus any custom monitors the game registers. Returns: latest values, and a history of the requested monitors - a rising orphan count is the usual sign of a leak. |
| runtime_screenshotA | Capture what a windowed game is drawing. Returns: the image, with its size. Needs a session started with windowed=true; a headless game has no pixels. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
TDQS
Scored across 49 tools
Two large families overlap heavily: debug_run/debug_output/debug_pause/debug_step_over/debug_inspect/debug_evaluate/debug_terminate mirror runtime_run/runtime_output/runtime_pause/runtime_step/runtime_inspect/runtime_eval/runtime_stop, and gdscript_references vs gdscript_references_in_file vs gdscript_references_batch, or gdscript_sync_file(s) vs gdscript_diagnostics, blur together. The descriptions do carefully explain the distinction (editor run slot vs separate process, breakpoint vs live, single vs batch), so an attentive agent can choose correctly, but the boundaries are not self-evident from names alone.
Names use a predictable snake_case namespace_verb/noun pattern with clean subsystem prefixes (gdscript_, debug_, runtime_, project_, scene_), and plural suffixes cleanly mark batch variants. Minor deviation: within families the second token mixes verbs and nouns (gdscript_complete vs gdscript_status vs gdscript_find), but this stays readable and consistent overall.
At 49 tools this is far past the comfortable range, and much of the bulk comes from parallel variants rather than distinct capabilities (three _batch wrappers, _sync_file vs _sync_files, references vs references_in_file, two near-identical run/inspect/output stacks). The breadth of the domain (LSP, debug adapter, standalone runtime, scene, project) explains some of it, but the surface is heavy enough that tool selection becomes a real burden.
Coverage is unusually strong end to end: navigation, hover, signature help, rename, validation-without-write, diagnostics, engine API lookup, project/scene validation, and both breakpoint debugging and live runtime introspection. Minor gaps remain, such as no workspace-wide symbol search and no hot-reload path for edited .tscn files or new assets (runtime_reload is GDScript-only), but these are workaround-able.