PrePrompt + Windsurf: Optimize Prompts Before They Reach the LLM

How PrePrompt Works in Windsurf

Windsurf implements the full Model Context Protocol, including the UserPromptSubmit hook. When you submit a prompt in Windsurf with PrePrompt installed:

1. The hook fires before the prompt reaches any LLM.
2. PrePrompt scores the prompt locally in under 1ms using its heuristic classifier — no API call.
3. Prompts scoring below 38 pass through unchanged with zero latency and zero cost.
4. Prompts scoring 38 or above are rewritten by Claude Haiku to be more specific and actionable.
5. Windsurf sends the optimized prompt to the LLM.
6. The interaction is logged locally to ~/.preprompt/history.db.

Installation

1. Install PrePrompt

   pip install preprompt

2. Run the setup wizard

   preprompt-install

   This saves your Anthropic API key and registers PrePrompt with every compatible IDE detected on your system, including Windsurf.

   To register Windsurf specifically:

   python scripts/install_windsurf.py

3. Verify the Windsurf config

   cat ~/.codeium/windsurf/mcp_config.json

   You should see a preprompt entry in the mcpServers object.

4. Restart Windsurf

   MCP servers are loaded at startup. Restart Windsurf to pick up the new configuration.

The MCP Configuration Entry

preprompt-install writes the following to ~/.codeium/windsurf/mcp_config.json:

{
  "mcpServers": {
    "preprompt": {
      "command": "python",
      "args": ["-m", "mcp_server.server"],
      "env": {
        "ANTHROPIC_API_KEY": "your-api-key-here"
      }
    }
  }
}

What Gets Optimized

• Pass — factual questions, very short prompts, conversational responses. Zero cost, zero latency.
• Enrich — complex prompts with vague action verbs, missing constraints, or multi-step requirements. Rewritten by Haiku (~1–2 seconds, ~$0.001).
• Clarify — very short vague prompts without technical context. Returns a clarifying question to help you write a better prompt.

Frequently Asked Questions