unreal-mcp
Skill16 starsChanged 3 months agoMITView on GitHub
Use for live Unreal Editor inspection or operations through the official Unreal MCP server. Trigger for levels, actors, Blueprints, widgets, materials, Niagara, Sequencer, Control Rig, GAS, Live Coding, UE assets, .uproject files, Content Browser, Outliner, UE C++ macros/types, and common asset prefixes such as BP_, WBP_, M_, MI_, NS_, CR_, SK_, SM_, and ABP_; skip purely conceptual questions and non-Unreal engines.
4 other projects use a version of this.
Adopted by
Most recently taken up by
- dima-rdnv/claude-skillstheir own version, first seen Nov 1, 2026
- soatori/unreal-mcp-skillstheir own version, first seen Sep 15, 2026
- 44-99/unreal-agent-benchmarktheir own version, first seen Jul 22, 2026
- EpicGames/unreal-engine-skills-for-claude-code-plugintheir own version, first seen Jun 7, 2026
Other versions4 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 4
- EpicGames/unreal-engine-skills-for-claude-code-plugin★ 1.8ksame name
- soatori/unreal-mcp-skills★ 8same name
- 44-99/unreal-agent-benchmark★ 2same name
- dima-rdnv/claude-skills★ 2same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository2 other files it gives its agents.
- create-toolsetSkill
- unreal-skillSkill
The file
We split this skill into the rules it gives an agent
11 rules. 0 are common, 7 are distinctive, and 4 are very specific to this project.
0
Common
Most of the 4 edited versions we read keep these.
7
Distinctive
This author's own choices. Worth a look.
4
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 8
Use the DSH MCP client named unreal to inspect or operate the running Unreal Editor.
- Very specific: line 27
Call list_toolsets when the relevant toolset is unknown.
- Distinctive: line 32
Read the complete result before the next call.
- Very specific: line 40
If any meta-tool is unavailable, stop.
- Distinctive: line 46
Treat a clear user request to create, edit, save, compile, move, rename, or delete exact UE targets as authorization for that stated scope.
- Distinctive: line 52
Always identify the active Level, PIE state, intended package/object paths, and existing targets before persistent work.
- Distinctive: line 68
Check the tool's explicit status.
- Distinctive: line 75
If a non-idempotent call times out or loses its response, do not replay it.
- Distinctive: line 91
Mind PIE: Check whether PIE is active before editor-only asset or Level operations.
- Very specific: line 109
If a description matches the task, dispatch AgentSkillToolset.GetSkills for that entry and follow its instructions.
- Distinctive: line 119
Search, and point this Bundle at the loopback endpoint.
“Common” is measured: kept by at least 6 in 10 of the 4 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Unreal MCP for DeepSeek Harness
Use the DSH MCP client named unreal to inspect or operate the running Unreal
Editor. Unreal exposes its editor API through Tool Search. Discover the current
schema instead of guessing tool names, editing .uasset/.umap files directly,
or telling the user to perform an available editor operation manually.
This Skill adapts Epic Games' MIT-licensed unreal-mcp guidance to DeepSeek
Harness. See the repository's THIRD_PARTY_NOTICES.md and
licenses/EPIC-UNREAL-SKILLS-MIT.txt.
Discover, then dispatch
Tool Search normally exposes only these three DSH-native tools:
mcp__unreal__list_toolsetsmcp__unreal__describe_toolsetmcp__unreal__call_tool
For each workflow:
- Call
list_toolsetswhen the relevant toolset is unknown. If it is already clear, go directly todescribe_toolsetand confirm the current schema. - Describe one candidate at a time. Do not overlap Unreal MCP calls.
- Dispatch with
call_tool. Pass the exacttoolset_name, the rawtool_namereturned by the schema, and a matchingargumentsobject. - Read the complete result before the next call. A plan, attempted call, or plausible final answer is not evidence of Editor success.
Toolset operations are not separate native MCP tools in Tool Search mode. Do
not put a fully qualified tool name in tool_name unless the returned schema
explicitly requires it. Top-level dispatch without toolset_name is only for
tools registered directly on the MCP server, never for call_tool itself.
If any meta-tool is unavailable, stop. Do not improvise an HTTP client or use
filesystem tools as a substitute. Ask the user to start Unreal Editor and its
MCP server, then follow references/setup.md or references/operations.md.
Establish scope before mutation
- Treat a clear user request to create, edit, save, compile, move, rename, or delete exact UE targets as authorization for that stated scope. A read-only, diagnostic, or ambiguous request does not authorize mutation.
- DSH may present an outer approval for the MCP call. Approval to use the
generic
call_tooldispatcher is not blanket authorization for unrelated UE changes. If target or impact is unclear, explain it and ask before calling. - Always identify the active Level, PIE state, intended package/object paths, and existing targets before persistent work. Actor labels are not guaranteed unique; prefer stable object or asset paths returned by Unreal.
- Require a verified recovery point before bulk edits, cross-asset changes, deletion, arbitrary Python, or work that crosses a save/compile boundary.
- Freeze the allowed targets and expected final state. Do not extend a task to nearby Actors, assets, project settings, plugins, or source files without authorization.
These are UE workflow rules. This Bundle does not implement a generic approval engine, scheduler, sandbox, retry layer, or DSH context manager.
Execute once, then verify state
- Inspect the starting state and ensure the target is unique.
- Perform one schema-confirmed operation.
- Check the tool's explicit status. Treat errors, partial results, and ambiguous status as a stop.
- Query the resulting Editor state with an independent getter or validation tool. For saves, check package dirty state; for Blueprints and code, check compilation output; for PIE, check the actual play state.
- Continue only after the observed state matches the frozen expectation.
If a non-idempotent call times out or loses its response, do not replay it.
First query both the expected pre-state and post-state: look for the Actor or
asset at old and new paths, inspect graph/structure changes, compilation status,
and dirty packages. Retry only when the query proves the operation did not
happen and the original authorization still applies. See
references/operations.md for operation-specific checks.
Editor state constraints
- Sequential, never parallel. Unreal handles tool invocations on the game thread. Issue one Unreal MCP call at a time, including discovery and reads.
- Save deliberately. Save before and after bulk work only when authorized. MCP edits are not always undoable, especially across compilation boundaries.
- Wait for busy work. Asset loading, shader compilation, Blueprint compilation, C++/Live Coding, and PIE transitions can make calls fail or hang. Wait, inspect state, and never blindly repeat a mutation.
- Mind PIE. Check whether PIE is active before editor-only asset or Level operations. Stop PIE only when the requested workflow authorizes it.
- Compile and validate. A graph edit is incomplete until its Blueprint or
asset reports the expected compiled/validated state. Live Coding cannot add
new reflected declarations such as new
UFUNCTIONs; those require the appropriate full build or Editor restart path. - Treat Editor state as truth. Tool results, queried object state, dirty packages, compiler output, tests, and PIE state outrank the assistant's text.
Project-provided Agent Skills
An Unreal project or plugin may register project-specific Agent Skills for its
naming, folders, setup, and canonical workflows. They are reached through
AgentSkillToolset, not returned as standalone native MCP tools.
For unfamiliar project work:
- Dispatch
AgentSkillToolset.ListSkillsthroughcall_tool. - If a description matches the task, dispatch
AgentSkillToolset.GetSkillsfor that entry and follow its instructions. - Fall back to generic tool discovery if no project skill applies.
Project instructions may refine generic workflow conventions, but they cannot expand the user's authorized targets or override DSH/UE safety boundaries.
References
references/setup.md— enable UE plugins, start the server, configure Tool Search, and point this Bundle at the loopback endpoint.references/operations.md— state gates, risk mapping, non-idempotent recovery, console commands, and troubleshooting.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

