New: watch your coding agents work, step by step, on a live board. Try the demo, no account

MCP token counter: what your tool list costs an agent

Every tool name, description and input schema an MCP server exposes can land in the agent's context before it does any work. Paste a tools/list response to see the cost per tool, the total, and what you can trim. No sign-up; what you paste stays on this page.

curl -s https://example.com/mcp \
  -H 'Content-Type: application/json' \
  -H 'Accept: application/json, text/event-stream' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

Add an Authorization header if your server needs a key. A server that insists on a session needs an initialize call first; any JSON with a tools array works here.

How the count works

The total is the length of the tool list as JSON divided by four, a common estimate for English and code. Each model has its own tokenizer and each client wraps tools its own way, so read it as a size to compare and track, not an exact bill. The budgets (about 5,000 tokens a server, 800 a tool, eight input fields) are rules of thumb, not limits any client enforces.

When the cost applies

Claude Code defers tool definitions through tool search by default, loading only names and server instructions at session start, then the full definition of each tool Claude looks up. It loads tools upfront when tool search is off or unsupported, and cuts a description loaded that way at 4,096 characters. Other clients differ, so check yours; in Claude Code, /context shows what is using your context window.

What we found on our own server

Taskhold's server has 22 tools and measured about 8,000 tokens. zod's .uuid() had written a 180-character regex into all 20 id fields. Swapping it for a custom check kept the same validation and took the list to about 7,200, without changing a word of a description. That is still over the rule of thumb: we kept the descriptions that tell agents how to report progress and ask you questions, and a size test in our build stops the list from growing.

Run a lot of agents?

Taskhold is a task list your agents report into over MCP: steps, progress and blocked questions on a live board you can check from your phone. $12/month or $120/year after the trial. Card up front. Cancel before day 7 and you pay nothing.