// MCP TOKEN COST

The GitHub MCP server costs 14,406 tokens before the first call.

Connect GitHub and its 86 tool definitions are loaded into the model's context on every request — 7.2% of a 200k window spent before your agent does anything.

QUICK ANSWER The GitHub MCP server's tool definitions consume 14,406 tokens — 13× the median MCP server (1,075 tokens). A scoped grant exposing only the tools you use cuts that roughly in proportion.

MEASURED FROM SCHEMAS 86 tools · 14,406 tokens · 7.2% of 200k · 1.4% 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 7.2%

1M WINDOW 1.4%

Corpus context: GitHub ranks #47 of 1,659 measured MCP servers by definition cost. The median is 1,075 tokens, p90 is 6,119, and the heaviest (Fusionauth) is 183,337 — 92% of a 200k window on its own.

// PER-TOOL BREAKDOWN

Where the 14,406 tokens go.

Each row is one tool definition as a tools/list entry — name, description and input schema — counted with o200k_base. Average: 168 tokens per tool.

Tool	Category	Tokens	% of server
actions_list	Read	569	3.9%
projects_write	Write	562	3.9%
pull_request_read	Read	430	3.0%
triage_issue	Write	377	2.6%
list_global_security_advisories	Read	360	2.5%
list_notifications	Read	357	2.5%
projects_list	Read	352	2.4%
issue_write	Write	345	2.4%
sub_issue_write	Write	315	2.2%
projects_get	Read	302	2.1%
add_comment_to_pending_review	Write	301	2.1%
list_issues	Read	285	2.0%
pull_request_review_write	Write	281	2.0%
actions_get	Read	253	1.8%
search_pull_requests	Read	253	1.8%
search_issues	Read	249	1.7%
search_repositories	Read	244	1.7%
get_job_logs	Read	240	1.7%
actions_run_trigger	Execute	235	1.6%
create_or_update_file	Write	235	1.6%
create_pull_request_with_copilot	Write	231	1.6%
assign_copilot_to_issue	Write	230	1.6%
list_pull_requests	Read	226	1.6%
issue_read	Write	223	1.5%
run_secret_scanning	Execute	219	1.5%
search_code	Read	213	1.5%
list_discussions	Read	210	1.5%
label_write	Write	209	1.5%
list_commits	Read	207	1.4%
update_pull_request	Write	193	1.3%
search_users	Read	192	1.3%
search_orgs	Read	187	1.3%
list_starred_repositories	Read	176	1.2%
get_copilot_space	Read	174	1.2%
list_code_scanning_alerts	Read	171	1.2%
get_repository_tree	Read	170	1.2%
list_secret_scanning_alerts	Read	161	1.1%
get_commit	Read	158	1.1%
push_files	Write	150	1.0%
get_file_contents	Read	148	1.0%
create_pull_request	Write	135	0.9%
get_copilot_job_status	Read	133	0.9%
list_repository_security_advisories	Read	129	0.9%
list_dependabot_alerts	Read	126	0.9%
list_gists	Read	126	0.9%
get_discussion_comments	Read	124	0.9%
merge_pull_request	Write	122	0.8%
github_support_docs_search	Read	120	0.8%
add_issue_comment	Write	119	0.8%
add_reply_to_pull_request_comment	Write	119	0.8%
list_org_repository_security_advisories	Read	113	0.8%
create_repository	Write	108	0.7%
mark_all_notifications_read	Write	108	0.7%
list_branches	Read	106	0.7%
list_releases	Read	106	0.7%
list_tags	Read	106	0.7%
manage_repository_notification_subscription	Write	104	0.7%
delete_file	Destructive	99	0.7%
update_pull_request_branch	Write	95	0.7%
create_gist	Write	90	0.6%
request_copilot_review	Read	87	0.6%
create_branch	Write	84	0.6%
manage_notification_subscription	Write	84	0.6%
update_gist	Write	84	0.6%
get_code_scanning_alert	Read	81	0.6%
get_release_by_tag	Read	81	0.6%
get_secret_scanning_alert	Read	81	0.6%
get_dependabot_alert	Read	80	0.6%
dismiss_notification	Read	73	0.5%
list_discussion_categories	Read	72	0.5%
get_notification_details	Read	69	0.5%
get_team_members	Read	69	0.5%
fork_repository	Read	67	0.5%
get_label	Read	67	0.5%
get_tag	Read	66	0.5%
get_teams	Read	63	0.4%
list_label	Read	63	0.4%
get_discussion	Read	62	0.4%
get_global_security_advisory	Read	56	0.4%
get_me	Read	55	0.4%
get_latest_release	Read	52	0.4%
unstar_repository	Read	49	0.3%
star_repository	Read	47	0.3%
get_gist	Read	46	0.3%
list_issue_types	Read	45	0.3%
list_copilot_spaces	Read	42	0.3%

// SCOPED GRANT

Most agents use a handful of these tools. They pay for all 86.

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 (168 tokens each).

Grant scope	Definition cost	Reduction
All 86 tools (no gateway)	14,406 tokens	—
3 granted tools	~503 tokens	−97%
5 granted tools	~838 tokens	−94%
10 granted tools	~1,675 tokens	−88%

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

//FAQ

GitHub token-cost questions.

How many tokens does the GitHub MCP server use?+

Its 86 tool definitions total 14,406 tokens — 7.2% 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 GitHub 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 GitHub's token usage?+

Expose fewer tools. A PolicyLayer grant scopes GitHub 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 503 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 05-06-2026 from the PolicyLayer scan database over all 86 catalogued GitHub tools. Counts refresh with every site build.

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

A PolicyLayer grant scopes GitHub 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.