How many tokens does your MCP server cost per request?
Every tool definition a server lists rides along in each model request, whatever the user asks. Name a server HISTOR observes, or paste its tools/list: you get its tokens, the money per 1,000 requests, the tools that weigh most and a README badge.
Free · no account · no wallet
What it checks
- The tokens a client puts in every model request for this server: each tool's name, description and input schema, counted with o200k_base.
- Where that sits among the 18,204 packages and 16,170 remote servers in HISTOR's snapshot of 19 October 2026.
- What it costs per 1,000 requests at a small and a mid-size model's input price, and read from a prompt cache.
- The heaviest tools, so you know which descriptions and schemas to trim first.
- A README badge for a server HISTOR observes.
What it does not check
One tokenizer: other models count differently, typically by tens of percent. The model provider's own tool-use text, and what a client drops or adds, are not counted. A lookup uses HISTOR's last observation and never contacts the server.
Privacy
A pasted list is counted in memory and answered; it is not stored. A lookup asks HISTOR for its last observation; the server itself is never contacted.
Show the count in your README
For a server HISTOR observes, the badge reads the current count, refreshed hourly. Put your server in the server parameter: an https endpoint, npm:<package>, pypi:<package> or a registry name, URL-encoded. A lookup on this page prints the exact line.
AI agents call it free
Free for agents too, at a lower rate limit:
POST https://modelmarket.dev/a2mcp/tool-list-costQuestions
Why does a tool list cost tokens on every request?
A model API keeps no memory between requests. A client that connects an MCP server sends that server's tool definitions with each request so the model can pick a tool, whether or not the question needs one.
How do I make my server's list cheaper?
Descriptions that say what a tool does and when to use it, and nothing more; schemas without repeated prose; fewer tools, or related ones behind a single tool. A client that keeps the list identical and early in the prompt can read it from a prompt cache at about a twentieth of the price.
Where do the comparison numbers come from?
From HISTOR, a public transparency log that keeps the current tool list of every MCP server it observes. On 19 October 2026 the median package listed 1,272 tokens and the 90th percentile 8,361. The method and the summary are published with it.
Other free checks
x402 payment check
Why a USDC transferWithAuthorization reverts: the signer, the token domain, the window and the seller's terms — and the domain a failing signature was really made for.
MCP rug-pull diff
Two tools/list results of one server — the one you approved and today's. A verdict, a word diff per tool, and a signal for each risky thing the change added.
MCP tool scan (WARDEN)
One tools/list. WARDEN, the open-source MCP firewall, checks every name, description, schema and annotation for prompt injection, exfiltration and credential theft.
MCP server history (HISTOR)
A server's address or registry name. When HISTOR first pinned its tool set, how often it changed, and what the pattern scan found — from a public, signed log.