> ## Documentation Index
> Fetch the complete documentation index at: https://docs.mainwp.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Optimize Token Usage

> Shrink the MCP server's footprint in your AI client's context window with compact schemas and tool filtering.

The full tool catalog costs roughly 13,000 tokens of your AI client's context window in standard mode (measured against a 62-tool catalog; the exact count varies by Dashboard version). That is fine for interactive use, but for long sessions or focused automation you can cut it substantially.

## Compact schema mode

Compact mode truncates tool descriptions to 60 characters and removes examples, trading roughly 20% of the catalog's token cost for less inline documentation:

```json theme={null}
{
  "schemaVerbosity": "compact"
}
```

```bash theme={null}
MAINWP_SCHEMA_VERBOSITY=compact
```

<Warning>
  Compact mode removes inline safety warnings from tool descriptions and relies on MCP semantic annotations (`readOnlyHint`, `destructiveHint`) instead. Verify your AI client surfaces those annotations before using compact mode in a setup where destructive tools are exposed. The server-side [confirmation gate](/mcp-server/safety) still applies either way.
</Warning>

## Filter the tool list

Fewer tools means fewer tokens, and it is the bigger lever. The ready-made configurations in [Restrict Available Tools](/mcp-server/guides/restrict-tools) cut the catalog cost by 53% to 94% depending on how narrow you go.

## Stack both

A focused automation setup that combines compact schemas with a four-tool whitelist reduces catalog token usage by over 90%:

```json theme={null}
{
  "schemaVerbosity": "compact",
  "allowedTools": ["list_sites_v1", "get_site_v1", "list_updates_v1", "run_updates_v1"]
}
```

This suits pipelines where the AI performs a narrow, well-understood task and does not need the rest of the catalog documented in its context.