// MCP TOKEN COST

The Unreal MCP server costs 11,937 tokens before the first call.

Every request your agent makes carries every tool definition this server exposes — context your code, documents and conversation can't use, mostly for tools the agent never calls. You don't need them all in the window, and you don't have to pay for them.

QUICK ANSWER The Unreal MCP server's 127 tool definitions consume 11,937 tokens — 6.0% of a 200k context window, and 6.4× the median MCP server (1,862 tokens). A scoped grant exposing only the tools you use cuts that roughly in proportion.

MEASURED FROM SCHEMAS tiktoken o200k_base · rank #1834 of 12,722 measured servers · refreshed every build Method →

// CONTEXT WINDOW SHARE

What that costs before your agent starts working.

Tool definitions are overhead: they occupy context on every request and compete with your code, documents and conversation history for the same window.

200K WINDOW 6.0%

1M WINDOW 1.2%

Corpus context: Unreal ranks #1834 of 12,722 measured MCP servers by definition cost. The median is 1,862 tokens, p90 is 13,674, and the heaviest (Ainumbers Mcp Apps) is 442,047 — 221% of a 200k window on its own. New to this? See MCP token cost and context window in the glossary.

Token cost is one axis. See the risk picture across your whole stack →

// PER-TOOL BREAKDOWN

Where the 11,937 tokens go.

Each row is one tool definition as a tools/list entry — name, description and input schema — counted with o200k_base. Average: 94 tokens per tool.

Tool	Category	Tokens	% of server
build_cook_run	Execute	284	2.4%
spawn_actor	Write	268	2.2%
add_graph_node	Write	220	1.8%
set_actor_transform	Write	200	1.7%
spawn_niagara_at_location	Execute	200	1.7%
spawn_blueprint_actor	Write	198	1.7%
connect_material_property	Write	196	1.6%
add_track	Write	159	1.3%
create_material_expression	Write	157	1.3%
connect_material_expressions	Write	156	1.3%
set_blueprint_property	Write	155	1.3%
set_niagara_color	Write	152	1.3%
set_material_instance_vector	Write	151	1.3%
build_target	Execute	149	1.2%
connect_graph_nodes	Write	149	1.2%
set_viewport_camera	Write	146	1.2%
set_niagara_vector	Write	136	1.1%
add_blueprint_variable	Write	133	1.1%
list_assets	Read	122	1.0%
duplicate_actors	Write	122	1.0%
build_graph	Execute	117	1.0%
set_actor_property	Write	116	1.0%
run_gauntlet	Execute	115	1.0%
add_blueprint_component	Write	115	1.0%
create_blueprint	Write	115	1.0%
cook_content	Execute	111	0.9%
import_asset	Write	109	0.9%
build_plugin	Execute	106	0.9%
package_project	Execute	106	0.9%
consolidate_assets	Write	105	0.9%
create_anim_blueprint	Write	105	0.9%
create_anim_montage	Write	105	0.9%
call_preset_function	Execute	104	0.9%
get_asset_references	Read	103	0.9%
apply_material	Write	103	0.9%
set_material_instance_texture	Write	103	0.9%
set_playback_range	Write	103	0.9%
create_material_instance	Write	102	0.9%
set_skeletal_mesh_lod	Write	101	0.8%
set_material_instance_scalar	Write	100	0.8%
set_niagara_bool	Write	100	0.8%
set_niagara_float	Write	100	0.8%
run_stat_command	Execute	99	0.8%
set_data_layer_state	Write	99	0.8%
set_preset_property	Write	98	0.8%
generate_collision	Write	96	0.8%
generate_lods	Execute	96	0.8%
sc_checkin	Write	95	0.8%
start_trace	Execute	94	0.8%
remove_graph_node	Destructive	93	0.8%
spawn_niagara_attached	Execute	93	0.8%
list_actors	Read	92	0.8%
set_streaming_source	Write	91	0.8%
execute_console_command	Execute	89	0.7%
add_actor_binding	Write	89	0.7%
delete_asset	Destructive	88	0.7%
get_preset_property	Read	88	0.7%
export_sequence_fbx	Write	87	0.7%
set_sequence_framerate	Write	87	0.7%
apply_anim_modifier	Write	86	0.7%
delete_material_expression	Destructive	85	0.7%
search_assets	Read	85	0.7%
add_blueprint_function	Write	85	0.7%
export_asset	Write	85	0.7%
set_actor_tags	Write	85	0.7%
create_level_sequence	Write	84	0.7%
render_sequence	Execute	84	0.7%
sc_status	Read	83	0.7%
create_material_function	Write	83	0.7%
execute_python	Execute	82	0.7%
duplicate_asset	Write	82	0.7%
run_automation_tests_by_category	Execute	80	0.7%
get_build_status	Read	80	0.7%
rename_asset	Write	80	0.7%
create_material	Write	79	0.7%
get_asset_info	Read	78	0.7%
sc_mark_for_add	Write	78	0.7%
take_screenshot	Read	77	0.6%
get_blueprint_info	Read	76	0.6%
get_anim_sequence_info	Read	75	0.6%
disable_plugin	Write	75	0.6%
enable_plugin	Write	75	0.6%
run_editor_utility_blueprint	Execute	74	0.6%
get_preset_info	Read	74	0.6%
get_sequence_info	Read	74	0.6%
sc_checkout	Write	73	0.6%
sc_revert	Write	73	0.6%
select_actors	Write	73	0.6%
reimport_skeletal_mesh	Write	72	0.6%
run_editor_utility_widget	Execute	71	0.6%
generate_lightmap_uvs	Write	71	0.6%
reinit_niagara_system	Execute	71	0.6%
resave_packages	Write	71	0.6%
delete_actor	Destructive	70	0.6%
run_automation_test	Execute	70	0.6%
list_graph_nodes	Read	70	0.6%
reset_niagara_system	Execute	69	0.6%
recompile_material	Execute	69	0.6%
start_csv_profiling	Execute	68	0.6%
get_undo_history	Read	67	0.6%
list_material_expressions	Read	67	0.6%
sc_diff	Read	67	0.6%
get_actor_properties	Read	66	0.6%
validate_assets	Execute	65	0.5%
compile_blueprint	Execute	64	0.5%
get_actor_components	Read	64	0.5%
validate_data	Execute	64	0.5%
save_asset	Write	64	0.5%
list_automation_tests	Read	62	0.5%
clean_project	Execute	56	0.5%
get_connection_status	Read	54	0.5%
fix_redirectors	Write	54	0.5%
content_audit	Execute	53	0.4%
generate_project_files	Write	51	0.4%
get_viewport_camera	Read	47	0.4%
list_data_layers	Read	47	0.4%
stop_csv_profiling	Write	46	0.4%
get_test_results	Read	46	0.4%
list_presets	Read	46	0.4%
run_map_check	Execute	45	0.4%
run_all_automation_tests	Execute	43	0.4%
stop_trace	Execute	43	0.4%
get_loaded_cells	Read	43	0.4%
list_plugins	Read	43	0.4%
save_all	Write	43	0.4%
redo	Write	41	0.3%
undo	Write	40	0.3%

// CUT THE COST

Your agent uses a handful of these tools. It pays for all 127.

You don't need all 127 of those definitions in the window. PolicyLayer is an MCP gateway that sits in front of Unreal: only the tools you grant are exposed to the agent, the rest never load. A smaller window means a sharper agent — less noise when it picks a tool — and every request costs less:

Grant scope	Definition cost	Reduction
All 127 tools (no gateway)	11,937 tokens	—
3 granted tools	~282 tokens	−98%
5 granted tools	~470 tokens	−96%
10 granted tools	~940 tokens	−92%

The risk dividend: 4 of these 127 tools are critical-risk (destructive or financial) and cost 336 tokens (3% of the definition load). Block them — the recommended starter policy — and you reclaim that context before tuning anything else.

Create a free account and register Unreal — nothing to install.
Grant only the tools you use — ungranted definitions never enter the context window.
Point your MCP client (Claude, Cursor, anything) at your gateway URL.

CUT UNREAL TOKEN COST →

Instant setup, no code required.

Model your own stack in the token-cost calculator, or see the Unreal policy for what a sensible grant looks like.

//FAQ

Unreal token-cost questions.

How many tokens does the Unreal MCP server use?+

Its 127 tool definitions total 11,937 tokens — 6.0% of a 200k context window — measured with tiktoken o200k_base over the serialised tools/list payload. Exact counts vary slightly by client and model.

Why does Unreal consume tokens before I send a message?+

MCP clients load every connected server's tool definitions — name, description, and input schema — into the model's context so it knows what it can call. That payload is charged against your context window on every request, whether or not a tool is used.

How do I reduce Unreal's token usage?+

Expose fewer tools. A PolicyLayer grant scopes Unreal to only the tools you allow — ungranted definitions are filtered out of the tool list, so they never enter the context window. A grant of 3 typical tools costs roughly 282 tokens, a 98% reduction.

Does deferred tool loading fix this?+

Partially, in some clients. Claude Code defers MCP tool schemas behind a tool-search step by default, and VS Code has experimental grouping — but you still pay tokens per search and reload, and Cursor, Windsurf and Gemini CLI load definitions upfront. Reducing the exposed tool set cuts the cost in every client.

// METHOD & REVIEW

How these numbers were measured.

01

Serialisation

Each tool is serialised as a tools/list entry — name, description, input schema — from the schemas in the PolicyLayer scan database. Clients differ slightly in framing, so treat counts as close estimates.

02

Tokeniser

tiktoken o200k_base (GPT-4o/o-series). Anthropic's current tokeniser isn't published, so Claude's exact counts will differ; for English text and JSON schemas the totals are close enough to treat these as estimates.

03

Deferred loading

Some clients now defer schema loading (Claude Code's tool search; VS Code experimental grouping). You still pay per search and reload — and Cursor, Windsurf and Gemini CLI load everything upfront.

Computed 20-10-2026 from the PolicyLayer scan database over all 127 catalogued Unreal tools. Counts refresh with every site build.