Install — zemt prompt-optimizer (MCP + skill)

Cost-route AI jobs before running them: classify → pick the cheapest capable model (15 models: Grok / Gemini / Claude) → get paste-ready optimized prompts. Free price/estimate tools; grok-4.3 analysis is token-gated. Prices updated July 2026.

⬇ MCP server (node, zero deps) ⬇ Skill (SKILL.md) GitHub (self-host)

Tools in the MCP

tool	what	auth
estimate_cost	price table for N jobs, budget math, all 15 models	free, no token
get_prices	current $/M token prices (July 2026)	free, no token
optimize_prompt	grok-4.3: classify, plan, route, paste-ready prompts	API token or self-host

Remote MCP endpoint (no download needed): https://ai.ze.mt/mcp — free tools work unauthenticated; add Authorization: Bearer <token> or use https://ai.ze.mt/mcp/<token> for optimize_prompt.

Claude Cowork (desktop)

Skill

Settings → Capabilities → add skill, or copy the folder:

mkdir "%USERPROFILE%\.claude\skills\prompt-optimizer" 2>nul
curl -o "%USERPROFILE%\.claude\skills\prompt-optimizer\SKILL.md" https://ai.ze.mt/download/prompt-optimizer-SKILL.md

MCP

Settings → Connectors → Add custom connector → URL:

https://ai.ze.mt/mcp            (free tools)
https://ai.ze.mt/mcp/<TOKEN>    (incl. optimize_prompt)

Claude Code (CLI)

# local stdio (download first):
curl -o zemt-optimizer-mcp.js https://ai.ze.mt/download/zemt-optimizer-mcp.js
claude mcp add --scope user zemt-optimizer -- node /full/path/zemt-optimizer-mcp.js

# or remote, no download:
claude mcp add --transport http zemt-optimizer https://ai.ze.mt/mcp

# skill (Windows):
curl -o "%USERPROFILE%\.claude\skills\prompt-optimizer\SKILL.md" --create-dirs https://ai.ze.mt/download/prompt-optimizer-SKILL.md
# skill (mac/linux):
curl -o ~/.claude/skills/prompt-optimizer/SKILL.md --create-dirs https://ai.ze.mt/download/prompt-optimizer-SKILL.md

Token for optimize_prompt: set env ZEMT_API_TOKEN or add --header "Authorization: Bearer <token>" on the http transport.

Grok Build

Zero config if you use Claude Code: Grok Build automatically reads Claude Code skills (~/.claude/skills/) and MCP servers. Install per the Claude Code section above and you are done — the skill appears as /prompt-optimizer.

Standalone: put the skill in ~/.grok/skills/prompt-optimizer/SKILL.md and add the MCP via /mcps in the TUI (command: node /path/zemt-optimizer-mcp.js).

Gemini CLI

Add to ~/.gemini/settings.json:

{
  "mcpServers": {
    "zemt-optimizer": {
      "command": "node",
      "args": ["/full/path/zemt-optimizer-mcp.js"],
      "env": { "ZEMT_API_TOKEN": "<optional>" }
    }
  }
}

Gemini has no Claude-style skills — paste the decision rules from the SKILL.md into your GEMINI.md context file instead.

REST API (no MCP needed)

# free:
curl https://ai.ze.mt/api/v1/prices
curl -X POST https://ai.ze.mt/api/v1/estimate -H "Content-Type: application/json" \
  -d '{"tokens_in":2000,"tokens_out":800,"jobs":290,"budget_usd":10}'

# token-gated:
curl -X POST https://ai.ze.mt/api/v1/optimize \
  -H "Authorization: Bearer <TOKEN>" -H "Content-Type: application/json" \
  -d '{"prompt":"...","targets":["claude"],"platform":"windows"}'

ai.ze.mt — prompt-optimizer MCP + skill