Free Tool

Your MCP Servers Have a
Hidden Token Tax

In clients that load every tool schema up front, every MCP server injects its full tool schema into every message you send. Hosts that defer MCP tools, Claude Code among them, load most schemas only when needed. That overhead compounds fast. Calculate yours.

72% of one team's context window was consumed by just 3 MCP servers Apideck, 2026
32× token overhead vs CLI for the same operation Scalekit, 2026
15-20K tokens of tool schemas loaded on every turn MindStudio, 2026
Calculator

Calculate Your MCP Token Tax

Select the MCP servers you run, adjust your usage, and see the real cost. Then see what jMRI structured retrieval does to that number.

Your Connected MCP Servers

80
6
$5.00
How we estimate: Token counts per server are based on published schema measurements. Each tool typically costs 550-1,400 tokens for its name, description, JSON schema, and parameter definitions. Counts are conservative estimates: your actual overhead may be higher if servers expose verbose descriptions. Sources: MindStudio, Apideck/DEV.to, The New Stack.

Your Token Tax

Schema overhead per turn
0
0 tokens × loaded every message (where all schemas load up front)
Monthly token waste
0
$0.00 / month burned on tool definitions
200K Context Window Usage 0% consumed by schemas
What's left for your actual conversation, code, and reasoning

With jMRI Structured Retrieval

Retrieval cost per query
~774
tokens: one search + one symbol fetch (median over the 15 benchmark task-runs; range 297 to 1,361). Full task-runs (search + 3 fetches) average ~1.5K tokens vs a 377K-token "read-everything" baseline (methodology). That's per query. jCodeMunch has a fixed cost of its own too: in clients that load every tool schema up front, its six-tool front door is 945 tokens of schemas (tiktoken cl100k_base, v1.108.388). Claude Code loads three of those tools in every request and adds jCodeMunch's server instructions to each session's first message.
Modeled cost of one retrieval call
$0
the call itself, at your price per million

Stop Paying the Token Tax

jCodeMunch, jDocMunch, and jDataMunch implement the jMRI specification: the open standard for token-efficient context retrieval. Instead of dumping entire files into your context window, retrieve exactly the symbol, section, or schema you need. Four operations. ~774 tokens per call, median (methodology).

Install jCodeMunch (Free) View Pricing Read the jMRI Spec