MCP overview

Prompt Builder

Build ready-to-paste prompts from any MCP client.

cuelara_build_prompt turns a rough idea into a complete, ready-to-paste prompt for a specific target model — clear and lean, grounded only in what the idea actually says, with bracketed placeholders for anything genuinely missing.

Fastest setup — Claude Code

bash
claude mcp add --transport http cuelara-token-optimizer https://cuelara.com/api/mcp

That's the whole server, so cuelara_compress_prompt (Token Optimizer) and cuelara_optimize_prompt (Prompt Optimizer) come along too. For other clients — Claude Desktop, Cursor, Copilot, Gemini CLI, Antigravity — see the client setup guides, which show the exact config for each one. This manual setup is the reliable path — prefer it.

How it gets triggered

You don't need to name the tool. Your assistant reads cuelara_build_prompt's description against what you just asked and decides on its own whether to call it — so anything that reads as "turn this idea into a finished prompt for a specific model" tends to trigger it, for example:

  • "Build me a prompt for Claude that does X."
  • "Write a ready-to-paste ChatGPT prompt for this idea."
  • "I need a prompt for Cursor that scopes this coding task properly."
  • "Turn this idea into a prompt I can hand to Gemini."

If it doesn't fire on its own (some assistants are more conservative about picking tools, or confuse it with cuelara_optimize_prompt), just name it directly: "Use cuelara_build_prompt on this."

Arguments

  • idea (string, required) — the rough idea to turn into a prompt (max 4,000 characters).
  • target (string, optional) — "Any AI model" (default), "ChatGPT", "Claude", "Gemini", "Grok", "DeepSeek", "Cursor / Windsurf", or "GitHub Copilot".
  • useCase (string, optional) — "General" (default), "Coding", "Writing", "Marketing", "Business", or "Research".
  • detail (string, optional) — "Concise", "Balanced" (default), or "Detailed".

Call it directly

bash
curl -X POST https://cuelara.com/api/mcp \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc": "2.0",
    "id": 1,
    "method": "tools/call",
    "params": {
      "name": "cuelara_build_prompt",
      "arguments": { "idea": "a chatbot that answers questions about our docs", "target": "Claude", "useCase": "Coding" }
    }
  }'

Add -H "Authorization: Bearer YOUR_TOKEN_HERE" (a personal token from /dashboard/mcp) to authenticate as yourself instead of anonymously — your own plan's daily limit applies, and paid-plan output skips the "Built by Cuelara.com" attribution.

Limits

Anonymous calls share the same free daily limit as the website's Prompt Builder, keyed by IP. Signed-in calls use your own account's plan limits instead.

Or set it up automatically

This only works if you paste it into an AI assistant running inside an editor or CLI with real file/shell tools and a workspace — Claude Code, Cursor, VS Code's Claude extension or Copilot Chat agent mode, Gemini CLI. It will not work in a plain chat window (claude.ai, the Claude desktop chat app) — those have no file access, so it can't write anything for you. If you're not sure which one you're in, use the manual steps above instead.

text
Connect the Cuelara MCP server to whichever AI coding tool you're running in right now (VS Code with the Claude extension, VS Code with GitHub Copilot, Cursor, Claude Desktop, Claude Code CLI, Gemini CLI, or Antigravity). Do this:


0. First check: do you actually have file-editing or shell-command tools available in this session right now (i.e. are you running as an agent inside an editor/CLI with a real workspace, not a plain chat window with no file access)? If you don't, say so plainly, ask me which tool I want to connect, and just give me the exact config JSON or command to paste in myself — don't guess or pretend to write a file you can't reach.
1. If you do have those tools, detect which tool/editor you're running inside. If you can't tell, ask me.
2. Ask me: "Do you want to authenticate with a personal Cuelara access token (higher daily limit, no attribution line on paid plans), or connect anonymously?" If I say yes, ask me to paste the token — I'll generate one at https://cuelara.com/dashboard/mcp.
3. Based on the detected tool, create or update the correct MCP config with:
   - Server name: cuelara
   - Server URL: https://cuelara.com/api/mcp
   - If I gave you a token, add header "Authorization: Bearer <token>"


   Use the right format and location for the tool:
   - VS Code (Claude extension or Copilot Chat agent mode): create/update .vscode/mcp.json in the current workspace, under a "servers" key, with "type": "http" and "url".
   - Cursor: create/update .cursor/mcp.json under a "mcpServers" key, with "url" (and "headers" if I gave a token).
   - Claude Desktop: edit claude_desktop_config.json (ask me my OS if you need the exact path) using the mcp-remote bridge — "command": "npx", "args": ["-y", "mcp-remote", "https://cuelara.com/api/mcp"] (append "--header", "Authorization: Bearer <token>" to args if I gave one).
   - Claude Code: don't write a file — instead run: claude mcp add --transport http cuelara https://cuelara.com/api/mcp (add --header "Authorization: Bearer <token>" if I gave one).
   - Gemini CLI: create/update ~/.gemini/settings.json under a "mcpServers" key, with "httpUrl" (and "headers" if I gave a token).
   - Antigravity: create/update its mcp_config.json under a "mcpServers" key, with "url" (and "headers" if I gave a token).
4. Show me the exact file content or command before writing or running it.
5. After it's in place, tell me how to verify the connection for that specific tool, and remind me that three tools become available: cuelara_compress_prompt (compress a verbose prompt), cuelara_optimize_prompt (turn a rough idea into a structured prompt), and cuelara_build_prompt (turn a rough idea into a ready-to-paste prompt for a specific target model).