// MCP TOKEN COST

The Local MCP server costs 8,009 tokens before the first call.

Connect Local and its 108 tool definitions are loaded into the model's context on every request — 4.0% of a 200k window spent before your agent does anything.

QUICK ANSWER The Local MCP server's tool definitions consume 8,009 tokens — 4.2× the median MCP server (1,905 tokens). A scoped grant exposing only the tools you use cuts that roughly in proportion.

MEASURED FROM SCHEMAS 108 tools · 8,009 tokens · 4.0% of 200k · 0.8% of 1M Method →

// CONTEXT WINDOW SHARE

What that buys 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 4.0%

1M WINDOW 0.8%

Corpus context: Local ranks #293 of 3,213 measured MCP servers by definition cost. The median is 1,905 tokens, p90 is 7,952, and the heaviest (Fusionauth) is 183,337 — 92% of a 200k window on its own.

// PER-TOOL BREAKDOWN

Where the 8,009 tokens go.

Each row is one tool definition as a tools/list entry — name, description and input schema — counted with o200k_base. Average: 74 tokens per tool.

Tool	Category	Tokens	% of server
servicenow_update_incident	Write	221	2.8%
m365_create_event	Write	198	2.5%
submit_qa_report	Write	154	1.9%
outlook_calendar_create_event	Write	148	1.8%
teams_send_channel_message	Write	146	1.8%
m365_send_email	Write	138	1.7%
excel_create	Write	125	1.6%
nordvpn_servers	Read	122	1.5%
m365_reply_email	Write	117	1.5%
servicenow_add_comment	Write	117	1.5%
teams_send_message	Write	115	1.4%
servicenow_create_incident	Write	112	1.4%
word_create	Write	110	1.4%
m365_list_emails	Read	107	1.3%
excel_write_cell	Write	104	1.3%
onedrive_search_files	Read	102	1.3%
excel_read	Read	99	1.2%
whatsapp_send_file	Write	99	1.2%
stocks_get_chart	Read	98	1.2%
ppt_create	Write	98	1.2%
todo_create_task	Write	98	1.2%
explorer_search	Read	97	1.2%
outlook_forward_email	Read	96	1.2%
outlook_send_email	Write	96	1.2%
gdrive_search_files	Read	95	1.2%
m365_search_emails	Read	95	1.2%
connect_servicenow	Write	95	1.2%
outlook_list_emails	Read	93	1.2%
onedrive_read_file	Read	91	1.1%
outlook_move_email	Write	87	1.1%
outlook_flag_email	Read	84	1.0%
onedrive_set_scope	Write	84	1.0%
onedrive_write_file	Write	84	1.0%
word_append	Read	83	1.0%
whatsapp_send_message	Write	83	1.0%
outlook_save_attachment	Write	82	1.0%
m365_list_events	Read	80	1.0%
outlook_reply_email	Read	79	1.0%
pdf_read	Read	79	1.0%
report_problem	Read	79	1.0%
onedrive_move_file	Write	79	1.0%
m365_delete_event	Destructive	78	1.0%
outlook_reply_all	Read	78	1.0%
stocks_get_quote	Read	78	1.0%
outlook_create_folder	Write	78	1.0%
servicenow_search_incidents	Read	76	0.9%
search_m365_directory	Read	75	0.9%
m365_get_contact	Read	73	0.9%
onedrive_delete_file	Destructive	72	0.9%
todo_list_tasks	Read	72	0.9%
list_m365_people_insights	Read	70	0.9%
teams_read_chat_messages	Read	70	0.9%
whatsapp_read_messages	Read	69	0.9%
todo_complete_task	Write	69	0.9%
teams_read_channel_messages	Read	68	0.8%
gdrive_list_files	Read	67	0.8%
get_m365_person	Read	67	0.8%
outlook_search_emails	Read	66	0.8%
servicenow_search_kb	Read	66	0.8%
outlook_calendar_delete_event	Destructive	65	0.8%
servicenow_get_incident	Read	65	0.8%
slack_read_channel_messages	Read	64	0.8%
whatsapp_search_messages	Read	62	0.8%
slack_search_messages	Read	61	0.8%
m365_search_contacts	Read	60	0.7%
gdrive_read_file	Read	59	0.7%
m365_read_email	Read	59	0.7%
onedrive_file_info	Read	58	0.7%
connect_m365_account	Write	58	0.7%
gdrive_set_scope	Write	58	0.7%
m365_list_contacts	Read	57	0.7%
teams_list_chats	Read	56	0.7%
whatsapp_list_chats	Read	56	0.7%
run_qa	Execute	55	0.7%
outlook_read_email	Read	55	0.7%
servicenow_list_my_incidents	Read	55	0.7%
explorer_list	Read	54	0.7%
stocks_search_symbol	Read	54	0.7%
outlook_calendar_list_events	Read	52	0.6%
word_read	Read	52	0.6%
outlook_get_contact	Read	51	0.6%
ppt_read	Read	51	0.6%
outlook_contacts_search	Read	50	0.6%
echo	Read	49	0.6%
teams_list_channels	Read	49	0.6%
run_diagnostics	Execute	46	0.6%
daily_brief	Read	46	0.6%
gdrive_file_info	Read	46	0.6%
nordvpn_status	Read	46	0.6%
onedrive_list_files	Read	46	0.6%
outlook_contacts_list	Read	45	0.6%
slack_list_channels	Read	44	0.5%
lmcp_welcome	Read	43	0.5%
lmcp_state	Read	41	0.5%
nordvpn_diagnose	Execute	40	0.5%
update_self_diagnosis	Write	40	0.5%
get_config	Read	39	0.5%
slack_list_workspaces	Read	38	0.5%
outlook_diagnose	Read	37	0.5%
teams_list_teams	Read	37	0.5%
disconnect_m365_account	Write	36	0.4%
gdrive_root	Read	31	0.4%
onedrive_root	Read	31	0.4%
outlook_calendar_list_names	Read	31	0.4%
disconnect_servicenow	Write	31	0.4%
todo_list_lists	Read	30	0.4%
update_local_mcp	Write	30	0.4%
outlook_list_accounts	Read	29	0.4%

// SCOPED GRANT

Most agents use a handful of these tools. They pay for all 108.

A PolicyLayer grant exposes only the tools you allow — ungranted definitions are filtered out of the tool list, so they never enter the context window. Estimates below assume typical-weight tools (74 tokens each).

Grant scope	Definition cost	Reduction
All 108 tools (no gateway)	8,009 tokens	—
3 granted tools	~222 tokens	−97%
5 granted tools	~371 tokens	−95%
10 granted tools	~742 tokens	−91%

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

//FAQ

Local token-cost questions.

How many tokens does the Local MCP server use?+

Its 108 tool definitions total 8,009 tokens — 4.0% 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 Local 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 Local's token usage?+

Expose fewer tools. A PolicyLayer grant scopes Local 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 222 tokens, a 97% 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.

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.

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.

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 07-06-2026 from the PolicyLayer scan database over all 108 catalogued Local tools. Counts refresh with every site build.

Expose only the tools you use — the rest never enter your context.

A PolicyLayer grant scopes Local to the tools you actually allow. Ungranted definitions never load, and every call that does run is checked against policy first.

SCOPE YOUR GRANT →

Free to start. No card required.

4,600+ MCP servers and 31,000+ tools scanned and risk-classified.