Grimoire
Read-only MCP server that gives Claude live access to your UE5 project state — Blueprints, interfaces, variables, functions, event graphs, materials, structs, and Data Assets.
README
Grimoire — UE5 Context Bridge
Read-only MCP server that gives Claude live access to your UE5 project state — Blueprints, interfaces, variables, functions, event graphs, materials, structs, and Data Assets.
Claude → MCP Server (stdio) → IPC Bridge (TCP) → Unreal Host (in editor)
- MCP Server: Exposes tools to Claude, routes requests over TCP
- Unreal Host: Runs inside UE5 editor, executes queries via Unreal Python API
What This Unlocks
Grimoire gives Claude live read access to your UE5 project. Combined with other MCP servers, this enables workflows that weren't previously possible:
- Grimoire + Notion — Claude reads your Blueprint architecture live and writes design summaries, task cards, or system documentation directly into Notion
- Grimoire + GitHub — Claude inspects your Blueprint systems and opens issues, updates wikis, or writes PR descriptions grounded in actual code
- Grimoire + Slack — Claude reads a Blueprint and posts a technical summary to your team channel
- Grimoire alone — Claude reasons about your systems in context, catches bugs, suggests refactors, and answers architecture questions without you describing anything
The pattern is: Grimoire provides the UE5 context, other MCP servers act on it.
Prerequisites
- Python 3.10+
- UE5 project with Python Script Plugin enabled
- Json Blueprint Utilities plugin enabled (built-in, free, Epic Games)
- Claude Desktop or another MCP client
Installation
1. Install Python dependencies
pip install -r requirements.txt
2. Configure the project
Copy the example config and set your UE5 project path (the folder containing your .uproject file):
cp config.toml.example config.toml
On Windows (PowerShell): Copy-Item config.toml.example config.toml
Edit config.toml and set root under [project]. config.toml is gitignored — keep your real paths local only. See config.toml.example for all available keys.
3. Install the Unreal Host
The editor must be able to import ue5_host on its Python path. Two common approaches:
- Copy — Copy the entire
ue5_hostfolder into your project (e.g. paste it atYourProject/Content/Python/ue5_host/). No symlink required; this is the simplest option on Windows. - Symlink — Point
Content/Python/ue5_hostat a single checkout elsewhere if you prefer not to duplicate files.
Target layout inside your UE5 project:
YourProject/
└── Content/
└── Python/
└── ue5_host/
├── ue5_host.py
├── handlers.py
└── __init__.py
You can also skip copy/symlink and pass a full absolute path to the startup script instead (see step 4).
4. Enable the Host in UE5
- Open your UE5 project
- Edit → Project Settings → Plugins → Python
- Under Startup Scripts, add:
ue5_host.ue5_host- Or the full path, e.g.
C:/path/to/grimoire-ue5/ue5_host/ue5_host
- Or the full path, e.g.
- Restart the editor (or run the script manually once)
5. Register the MCP Server with Claude Desktop
Edit your Claude Desktop config (%APPDATA%\Claude\claude_desktop_config.json on Windows):
{
"mcpServers": {
"ue5-context": {
"command": "python",
"args": ["-m", "ue5_mcp.mcp_server"],
"cwd": "C:/path/to/grimoire-ue5",
"env": {
"UE5_MCP_CONFIG": "C:/path/to/grimoire-ue5/config.toml"
}
}
}
}
Replace cwd and UE5_MCP_CONFIG with your actual paths. Add other MCP servers (Notion, GitHub, Slack etc) to the same config to enable multi-server workflows.
Tools
| Tool | Description |
|---|---|
ping |
Check if the UE5 editor host is reachable |
list_blueprints |
List Blueprint assets (optional: path_prefix, name_substring) |
get_blueprint |
Full inspection: parent class, components, variables, functions, event graphs |
list_components |
Components on a Blueprint actor |
get_variables |
All variables with resolved types. include_locals=True for function-scope locals |
list_interfaces |
All Blueprint Interfaces in the project |
get_interface |
Full interface inspection: function signatures with typed input/output pins |
asset_search |
Search assets by class and name |
find_event_bindings |
Find Blueprints implementing a given event or interface (via cache) |
query_cache |
Query the SQLite cache by parent class, function name, variable name, or type reference |
get_data_asset |
Read property values from a Blueprint Data Asset instance (walk speeds, scalars, config) |
get_struct |
Inspect a UserDefinedStruct: field names and types |
get_material |
Inspect a Material or MaterialFunction: exposed parameters (scalar, vector, texture), output slot connections, and material function calls |
What Grimoire Can Read
Blueprints
- Parent class — resolved via Asset Registry
- Component hierarchy — full SCS tree
- Variables — names and resolved types (
EGait,FInventorySheet,BP_PlayerState_C— not rawByteProperty) - Functions — names, typed input/output pins, execution flow body
- Event graphs — bound lifecycle events (
ReceiveBeginPlay,ReceiveControllerChanged), custom events, and interface events — with full execution body where traceable - Warnings — honest flags when data is partially unavailable, with workarounds
Data Assets
- Property values — actual configured data from DA instances (
BaseWalkSpeed: 215,SprintMultiplier: 2.85) - Nested struct values — full property tree with GUID suffixes cleaned
Structs
- Field definitions — all
UserDefinedStructfields with resolved types - Navigable type system —
FInventorySheet → FInventoryContainer → FItemStackfully traversable
Materials
- Scalar parameters — name + default value (
HexScale: 50.0,StateBlend: 1.0) - Vector parameters — name + default RGBA (
OpenColor,LockedColor) - Texture parameters — name + default texture reference
- Output slot connections — which expression feeds
EmissiveColor,Opacityetc - Material function calls — which
MF_*functions are used
Cache
- SQLite persistence — parsed data survives across sessions
- Auto-invalidation — dirty flag watchdog marks stale entries every 15 seconds after asset saves, no manual cache clears needed
Execution Body Format
Function and event bodies reconstruct the logical spine of each graph as a readable step list:
call SubscribeToStats_Player
set CachedInvComp
bind_delegate(OnInventorySheetUpdate)
branch
macro:IsValid
dynamiccast
set BP_PlayerState
bind_delegate(OnStatsChanged)
return ValidationResult (SInteractionValidationResults) <- CurrentResult
select (bool) [False, True]
Node types surfaced: call, set, get, branch, macro:Name, bind_delegate(Name), dynamiccast, switchenum, array.op, select, make, break, return X (Type) <- Source
Known Limitations
- Interface implementation list — UE5 Python API does not expose which interfaces a Blueprint implements. Workaround: use
query_cache(tool='query_blueprints', has_function='FunctionName')to find implementors by shared function name. - Branch-aware execution bodies — function bodies are flat step lists; branch true/false paths are not nested. UE5's T3D export does not serialize exec pin connections. For complex branching functions, a Blueprint screenshot helps cross-reference execution paths.
- EventGraph partial traceability — events whose exec chain begins with a macro node (e.g.
Switch Has Authority) cannot be fully walked from T3D data. These events surface with anEVENTGRAPH_PARTIALwarning and[exec chain not traceable]body. The event's existence and binding are still reported correctly. - Deeply nested pure function chains — first-level pure node sources are resolved (
<- select,<- GetIsLocked); deeper chains may still show<- ?.
Both the interface list and exec pin connection gaps are known UE5 Python API limitations. An Epic support ticket has been filed requesting exec pin LinkedTo in T3D exports and BlueprintGeneratedClass.interfaces in Python bindings.
How It Works
The UE5 Python API restricts direct access to Blueprint internals. Grimoire uses three alternative pathways:
- SubobjectDataSubsystem —
k2_gather_subobject_data_for_blueprint()+export_text()parsing for components - JsonObjectGraphFunctionLibrary.stringify() — for variable names, resolved types (
PinSubCategoryObject→ actual type name), and Data Asset property values - AssetExportTask T3D export — two-pass node graph parsing for function signatures, execution flow, event graphs, macros, delegates, Select nodes, and Material parameters
Requires the Json Blueprint Utilities plugin (built-in, free).
Switching Projects
- Update
config.toml: change[project].rootand optionally[ipc].port - For multiple editors: set
UE5_MCP_PORTper project (e.g.65432,65433) - Restart Claude Desktop after config changes
Troubleshooting
Editor offline / Connection refused
- Ensure the UE5 editor is open with your project loaded
- Confirm the host started (check Output Log for
UE5 Context Bridge: listening on...) - Verify
config.tomlport matches the host (default65432)
Port conflict
- Use a different port in
config.tomland setUE5_MCP_PORTin environment - Or run only one UE5 editor at a time
Timeout
- Increase
timeout_secinconfig.toml - Large projects may need more time for asset scans
Python startup script not running
- Check that the Python Script Plugin is enabled
- Use the full absolute path to
ue5_hostin Startup Scripts if a relative path fails
Cache stale after code changes
- Clear specific entries:
DELETE FROM blueprint_cache WHERE asset_path LIKE '%BlueprintName%' - Cache DB location:
YourProject/Saved/Grimoire/cache.db - The watchdog auto-invalidates on asset save — manual clears only needed after handler code changes
Roadmap
V7 (planned)
- Animation Blueprint anim graph support
- Branch-aware body format — pending Epic Python API exposure of exec pin connections
- Multi-project switching without config changes
grimoire_setup.py— automated install and config generation
License & contributing
Grimoire is dual-licensed. AGPL-3.0 applies by default; see LICENSE for the full text and DUAL_LICENSE.md for commercial licensing.
Contributions, issues, and PR expectations: CONTRIBUTING.md.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.