Managing skills and MCP

The Manage tab turns off the skills and MCP servers that the audit flags as expensive, without editing config files by hand.

What it does

Skills and MCP servers each add token overhead to every session: skill metadata loads on start, and MCP servers contribute their full tool schema. The audit identifies the heaviest contributors. The Manage tab, and its CLI equivalents, let you disable a contributor in one action so it stops loading, then restore it later when you need it.

Disabling a skill moves it out of the skills directory to a backup location, so nothing is deleted. Disabling an MCP server removes its entry from settings.json. Both actions are reversible, and both apply to new tool calls without restarting the session.

When to use it

Use it after the audit shows a skill or MCP server costing more overhead than it earns you. Disable the ones you rarely use, keep the ones you rely on, and restore on demand.

Default state

The toggles are always available on every platform. Nothing is disabled until you disable it. There is no automatic toggling: Token Optimizer never turns a skill or server off on your behalf.

How to turn off

There is no background behavior to disable. The toggles act only when you run them. To reverse any toggle you made, restore the skill or re-enable the server:

cd ~/.claude/skills/token-optimizer/scripts
python3 measure.py skill restore SKILL_NAME
python3 measure.py mcp enable SERVER_NAME

Commands

Claude Code

cd ~/.claude/skills/token-optimizer/scripts

# Skills (archive moves it to backup, restore brings it back)
python3 measure.py skill archive SKILL_NAME
python3 measure.py skill restore SKILL_NAME

# MCP servers (disable removes from settings.json, enable restores it)
python3 measure.py mcp disable SERVER_NAME
python3 measure.py mcp enable SERVER_NAME

Codex

Codex keeps skills and MCP in different locations, so it has its own verbs.

TOKEN_OPTIMIZER_RUNTIME=codex python3 measure.py codex-skill disable NAME
TOKEN_OPTIMIZER_RUNTIME=codex python3 measure.py codex-skill enable NAME
TOKEN_OPTIMIZER_RUNTIME=codex python3 measure.py codex-mcp disable NAME
TOKEN_OPTIMIZER_RUNTIME=codex python3 measure.py codex-mcp enable NAME

codex-skill accepts an optional --path PATH when your Codex project is not at the default location.

From the dashboard

Open the dashboard, go to the Manage tab, and use the toggles. Each toggle runs the same command shown above.

python3 measure.py dashboard

Defaults and thresholds

Setting	Value
Skill disable mechanism	Move to backup directory (reversible)
MCP disable mechanism	Remove entry from `settings.json` (reversible)
Effect timing	New tool calls; no session restart needed
Automatic toggling	Never

Risk rating

Low. Disabling a skill or server changes what loads next session, and a tool you turned off will not be available until you restore it. Both actions are reversible with a single command, and skills are archived rather than deleted, so a mistaken toggle costs nothing but a restore.

None. The toggles act on files and settings.json directly. See the configuration reference for unrelated settings.

Platform availability

Skill and MCP toggles work on Claude Code (CLI and VS Code). Codex uses the codex-skill and codex-mcp verbs. Other platforms manage their own skill and connector inventories; check the capability matrix.

Setup audit: find the offenders to toggle.
The dashboard: the Manage tab UI.
CLI reference: every skill, mcp, codex-skill, and codex-mcp flag.