// MCP TOKEN COST

The Looking Glass MCP server costs 5,080 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 Looking Glass MCP server's 71 tool definitions consume 5,080 tokens — 2.5% of a 200k context window, and 2.6× the median MCP server (1,920 tokens). A scoped grant exposing only the tools you use cuts that roughly in proportion.

MEASURED FROM SCHEMAS tiktoken o200k_base · rank #2388 of 8,127 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 2.5%

1M WINDOW 0.5%

Corpus context: Looking Glass ranks #2388 of 8,127 measured MCP servers by definition cost. The median is 1,920 tokens, p90 is 12,644, and the heaviest (Ainumbers Mcp Apps) is 325,784 — 163% 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 5,080 tokens go.

Each row is one tool definition as a tools/list entry — name, description and input schema — counted with o200k_base. Average: 72 tokens per tool.

Tool	Category	Tokens	% of server
test_scenario_run	Execute	520	10.2%
test_auth_flow	Other	182	3.6%
browser_login	Execute	150	3.0%
test_assert	Other	135	2.7%
test_fill_form	Execute	134	2.6%
test_chaos	Execute	130	2.6%
test_watch_events	Other	127	2.5%
browser_mock_route	Execute	120	2.4%
browser_extract	Read	117	2.3%
browser_explore	Execute	103	2.0%
browser_set_cookie	Write	96	1.9%
browser_screenshot	Read	95	1.9%
browser_wait_for	Execute	94	1.9%
browser_fill_and_submit	Execute	94	1.9%
browser_scroll	Execute	94	1.9%
browser_network_requests	Read	86	1.7%
session_export_playwright	Other	86	1.7%
browser_wait_until_stable	Execute	82	1.6%
browser_diagnose	Other	78	1.5%
browser_drag	Execute	76	1.5%
vault_login	Other	75	1.5%
browser_type	Execute	74	1.5%
browser_smart_click	Execute	73	1.4%
browser_console_messages	Read	72	1.4%
browser_resolve_element	Execute	72	1.4%
browser_select_option	Execute	72	1.4%
browser_go	Execute	69	1.4%
browser_set_localstorage	Write	68	1.3%
browser_clear_storage	Destructive	67	1.3%
browser_press_key	Execute	67	1.3%
browser_tab_select	Execute	67	1.3%
session_replay	Read	64	1.3%
visual_baseline	Read	64	1.3%
vault_inject	Other	62	1.2%
browser_evaluate	Execute	60	1.2%
browser_action_history	Read	57	1.1%
browser_get_localstorage	Read	57	1.1%
browser_click	Execute	56	1.1%
browser_close	Destructive	56	1.1%
browser_navigate	Execute	55	1.1%
browser_hover	Execute	55	1.1%
visual_compare	Read	55	1.1%
vault_delete	Destructive	54	1.1%
browser_tab_new	Execute	52	1.0%
browser_get_cookies	Read	51	1.0%
session_start	Execute	48	0.9%
browser_performance_audit	Execute	36	0.7%
test_accessibility_audit	Read	36	0.7%
test_chaos_clear	Destructive	36	0.7%
browser_clear_mocks	Destructive	35	0.7%
browser_analyze_page	Read	35	0.7%
browser_clean_slate	Destructive	35	0.7%
browser_list_mocks	Read	35	0.7%
browser_suggest_actions	Other	35	0.7%
test_generate_assertions	Other	35	0.7%
test_scenario_status	Read	35	0.7%
browser_clear_history	Destructive	34	0.7%
browser_diff_state	Read	34	0.7%
browser_error_report	Read	34	0.7%
browser_go_back	Execute	34	0.7%
browser_go_forward	Execute	34	0.7%
browser_health_check	Read	34	0.7%
browser_recover	Execute	34	0.7%
browser_snapshot_state	Read	34	0.7%
browser_tab_list	Read	34	0.7%
browser_workflow	Execute	34	0.7%
test_stop_watch	Other	34	0.7%
browser_snapshot	Read	33	0.6%
session_end	Destructive	33	0.6%
session_list	Read	33	0.6%
vault_list	Read	33	0.6%

// CUT THE COST

Your agent uses a handful of these tools. It pays for all 71.

You don't need all 71 of those definitions in the window. PolicyLayer is an MCP gateway that sits in front of Looking Glass: 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 71 tools (no gateway)	5,080 tokens	—
3 granted tools	~215 tokens	−96%
5 granted tools	~358 tokens	−93%
10 granted tools	~715 tokens	−86%

The risk dividend: 8 of these 71 tools are critical-risk (destructive or financial) and cost 350 tokens (7% 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 Looking Glass — 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 LOOKING GLASS TOKEN COST →

Instant setup, no code required.

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

//FAQ

Looking Glass token-cost questions.

How many tokens does the Looking Glass MCP server use?+

Its 71 tool definitions total 5,080 tokens — 2.5% 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 Looking Glass 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 Looking Glass's token usage?+

Expose fewer tools. A PolicyLayer grant scopes Looking Glass 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 215 tokens, a 96% 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 05-09-2026 from the PolicyLayer scan database over all 71 catalogued Looking Glass tools. Counts refresh with every site build.