The hidden tax of MCP
gawk at an MCP server before you trust it.
You pay its tax on every prompt — before you type a word.
Connect an MCP server and it loads every tool definition into your model — used or not, on every request.
We measured the ones developers actually use.
The two most-installed dev-tool MCPs load 35,600 tokens at connect — roughly a quarter of your context, gone before your first prompt.
The surprise: it isn't how many tools. It's how heavy. The heaviest packs 20 tools into 23,085 tokens; Cloudflare fits 23 into 3,570 — 6.5× lighter, same job.
| Server | Tools | Tokens at connect |
|---|---|---|
| Heaviest dev-tool MCP | 20 | 23,085 |
| The most-installed MCP | 44 | 12,511 |
| Cloudflare | 23 | 3,570 |
| Exa | 2 | 520 |
Measured locally with mcpgawk. Heavy servers anonymised; lean ones named with credit. Nothing uploaded.
And heavy isn't just expensive.
In controlled tests, once the tool list gets long the model picks the wrong tool more often — one study: 13.6% right, versus 43.1% when the list is trimmed. The fix already exists: Cloudflare and Exa prove lean is a choice.
See your own tax.
One command. Nothing leaves your machine.
pip install mcpgawk mcpgawk scan mcp.jsonlocal · no inventory ever leaves your machine