// MCP TOKEN COST

The Meridian MCP server costs 67,711 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 Meridian MCP server's 175 tool definitions consume 67,711 tokens — 34% of a 200k context window, and 33× the median MCP server (2,038 tokens). A scoped grant exposing only the tools you use cuts that roughly in proportion.

MEASURED FROM SCHEMAS tiktoken o200k_base · rank #17 of 6,689 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 34%

1M WINDOW 6.8%

Corpus context: Meridian ranks #17 of 6,689 measured MCP servers by definition cost. The median is 2,038 tokens, p90 is 12,122, and the heaviest (Ainumbers Mcp Apps) is 269,012 — 135% 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 67,711 tokens go.

Each row is one tool definition as a tools/list entry — name, description and input schema — counted with o200k_base. Average: 387 tokens per tool.

Tool	Category	Tokens	% of server
generate_handoff	Write	2,636	3.9%
update_sprint_item	Write	2,583	3.8%
add_sprint_item	Write	2,214	3.3%
add_sprint_item_pointer	Write	1,803	2.7%
complete_sprint_item	Write	1,493	2.2%
update_paragraph	Write	1,150	1.7%
execute_batch	Execute	1,093	1.6%
record_handoff_correction	Destructive	962	1.4%
fan_out_sprint_items	Destructive	917	1.4%
claim_sprint_item	Destructive	869	1.3%
commit_proposal_promotion	Write	848	1.3%
batch_mutate	Destructive	809	1.2%
start_session	Execute	801	1.2%
configure_wave_gate	Write	795	1.2%
prospect_symbol	Execute	787	1.2%
batch_read	Read	758	1.1%
save_profile_layer	Write	676	1.0%
update_workspace_settings	Write	675	1.0%
add_custom_hook	Write	670	1.0%
set_executor_config	Write	669	1.0%
request_hitl	Destructive	643	0.9%
insert_equation	Write	640	0.9%
search_outputs	Read	638	0.9%
complete_wave_gate	Execute	631	0.9%
set_capability_profile	Write	610	0.9%
link_flag_to_section	Write	603	0.9%
resume_wave	Read	588	0.9%
update_custom_hook	Write	558	0.8%
add_note	Write	539	0.8%
index_table	Write	533	0.8%
start_wave_run	Write	532	0.8%
preview_proposal_promotion	Read	529	0.8%
resolve_sprint_item_pointers	Write	519	0.8%
search_code_semantic	Read	511	0.8%
verify_handoff_token	Read	511	0.8%
ingest_document	Write	510	0.8%
annotate_outputs	Write	502	0.7%
index_figure	Write	492	0.7%
index_equation	Write	479	0.7%
analyze_model_efficiency	Read	458	0.7%
claim_file	Write	453	0.7%
get_flag_drift	Read	450	0.7%
link_figure_caption	Write	442	0.7%
finalize_wave_run	Write	440	0.6%
find_similar_figure	Read	440	0.6%
get_notes	Read	440	0.6%
link_table_caption	Write	436	0.6%
ingest_document_structure	Write	429	0.6%
set_capability_manifest	Write	415	0.6%
get_effective_capability_profile	Read	411	0.6%
resolve_citations	Write	409	0.6%
set_active_repo	Write	404	0.6%
find_symbol_usages	Read	395	0.6%
assign_sprint_waves	Write	391	0.6%
get_server_logs	Read	388	0.6%
update_md_section	Write	387	0.6%
find_outputs_by_source	Read	385	0.6%
search_server_logs	Read	385	0.6%
get_effective_profile	Read	382	0.6%
merge_project	Write	378	0.6%
clone_profile_layer	Destructive	371	0.5%
link_manual_github_issue	Write	361	0.5%
claim_docx_region	Write	358	0.5%
save_finding	Write	351	0.5%
add_subtask	Write	350	0.5%
purge_ai_log	Destructive	347	0.5%
get_latex_structure	Read	344	0.5%
get_workspace_proposals	Read	344	0.5%
get_citation_edges	Read	341	0.5%
request_manual_issue_screening_toggle	Write	335	0.5%
set_parent_project	Write	331	0.5%
update_decision	Write	327	0.5%
activate_profile_layer	Write	325	0.5%
validate_assumption	Write	313	0.5%
pin_decision	Write	309	0.5%
save_blog_post	Write	308	0.5%
run_verification	Execute	307	0.5%
get_server_log_checkpoint	Read	306	0.5%
add_insight	Write	303	0.4%
get_parallelizable_groups	Read	302	0.4%
reset_plugin_override	Destructive	301	0.4%
store_finding	Write	297	0.4%
github_search	Read	295	0.4%
add_sprint_note	Write	288	0.4%
promote_proposal	Write	287	0.4%
capture_research_finding	Write	285	0.4%
paper_search	Read	283	0.4%
get_tunnel_diagnostics	Read	281	0.4%
get_planning_brief	Read	279	0.4%
find_similar_equation	Read	277	0.4%
analyze_sprint	Read	273	0.4%
get_flag_registry	Read	273	0.4%
export_ai_log	Write	273	0.4%
export_ai_log_artifacts	Write	270	0.4%
get_session_brief	Read	269	0.4%
find_similar_table	Read	267	0.4%
get_sprint_items	Read	265	0.4%
reset_profile_layer	Destructive	260	0.4%
send_message	Write	260	0.4%
list_hitl_requests	Read	258	0.4%
update_workspace_sprint_item	Write	257	0.4%
add_workspace_sprint_item	Write	256	0.4%
delete_custom_hook	Destructive	252	0.4%
log_task	Write	250	0.4%
add_workspace_proposal	Write	248	0.4%
get_sprint_progress	Read	247	0.4%
get_capability_manifest	Read	244	0.4%
register_session	Write	238	0.4%
set_sprint	Write	236	0.3%
refresh_context	Read	235	0.3%
split_sprint_item	Write	234	0.3%
create_project	Write	234	0.3%
social_search	Read	233	0.3%
set_agent_instructions	Write	233	0.3%
get_document_structure	Read	229	0.3%
merge_sprint_items	Write	228	0.3%
clear_capability_profile	Destructive	227	0.3%
release_docx_region_claims	Destructive	224	0.3%
list_profile_layers	Read	224	0.3%
reconcile_sprint_drift	Read	224	0.3%
get_findings	Read	222	0.3%
get_profile_layer	Read	219	0.3%
advance_proposal_status	Write	218	0.3%
checkpoint	Write	212	0.3%
search_all	Read	212	0.3%
delete_sprint_item_pointer	Destructive	210	0.3%
idle_until_all_done	Read	210	0.3%
get_context_block	Read	209	0.3%
refresh_tool_manifest	Read	209	0.3%
get_connection_log	Read	208	0.3%
receive_messages	Read	202	0.3%
get_file_claims	Read	199	0.3%
idle_until_session_done	Read	197	0.3%
snapshot_graph_metrics	Write	196	0.3%
load_handoff	Read	194	0.3%
set_north_star	Write	194	0.3%
get_workspace_sprint_items	Read	193	0.3%
rename_project	Write	193	0.3%
add_workspace_note	Write	191	0.3%
pin_workspace_decision	Write	188	0.3%
search_synthesis	Read	181	0.3%
answer_hitl	Write	174	0.3%
get_docx_region_claims	Read	173	0.3%
get_sprint_item_pointers	Read	173	0.3%
get_custom_hooks	Read	168	0.2%
set_goal	Write	165	0.2%
archive_decision	Destructive	164	0.2%
read_note	Read	151	0.2%
get_profile_layer_revisions	Read	150	0.2%
dismiss_hitl	Write	148	0.2%
release_file	Write	142	0.2%
get_session_log	Read	142	0.2%
complete_workspace_sprint_item	Write	140	0.2%
get_insights	Read	134	0.2%
get_session_activity	Read	133	0.2%
get_pinned_decisions	Read	131	0.2%
delete_note	Destructive	126	0.2%
list_sessions	Read	120	0.2%
get_goal	Read	118	0.2%
get_plugin_details	Read	116	0.2%
search_tasks	Read	113	0.2%
get_agent_instructions	Read	110	0.2%
get_sprint_notes	Read	110	0.2%
get_graph_diff	Read	108	0.2%
list_plugins	Read	100	0.1%
get_project_by_name	Read	92	0.1%
get_blog_posts	Read	83	0.1%
get_tasks	Read	83	0.1%
list_projects	Read	82	0.1%
get_symbol_hotspots	Read	81	0.1%
get_hitl_request	Read	70	0.1%
get_workspace_settings	Read	64	0.1%
get_symbol_claims	Read	61	0.1%
get_workspace_decisions	Read	56	0.1%
get_workspace_notes	Read	52	0.1%

// CUT THE COST

Your agent uses a handful of these tools. It pays for all 175.

You don't need all 175 of those definitions in the window. PolicyLayer is an MCP gateway that sits in front of Meridian: 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 175 tools (no gateway)	67,711 tokens	—
3 granted tools	~1,161 tokens	−98%
5 granted tools	~1,935 tokens	−97%
10 granted tools	~3,869 tokens	−94%

The risk dividend: 15 of these 175 tools are critical-risk (destructive or financial) and cost 6,682 tokens (10% 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 Meridian — 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 MERIDIAN TOKEN COST →

Instant setup, no code required.

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

//FAQ

Meridian token-cost questions.

How many tokens does the Meridian MCP server use?+

Its 175 tool definitions total 67,711 tokens — 34% 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 Meridian 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 Meridian's token usage?+

Expose fewer tools. A PolicyLayer grant scopes Meridian 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 1,161 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 17-08-2026 from the PolicyLayer scan database over all 175 catalogued Meridian tools. Counts refresh with every site build.