Developers
One engine. Every surface.
The diagnosis, optimization and security scan behind the web app are available over REST, as an MCP server that works in any MCP-capable AI client, and inside ChatGPT, Claude and Gemini through the Chrome extension.
Pilot status: all surfaces run the same rule-based engine (no AI model yet). The MCP server runs entirely on your machine and is downloadable below. API keys are issued to pilot customers on request.
Developer API
Same JSON shapes as the engine, versioned under /v1. Use it before sending prompts to your own model calls, or to gate agent instructions in CI.
POST
/v1/analyzeDiagnosis only: task, intent, ranked missing context and questions, constraints, assumptions, security, scores.
POST
/v1/optimizeOptimized prompt, change notes, intent-preservation checks, before/after scores, labelled assumptions.
POST
/v1/security/scanSecrets, PII, injection, exfiltration and agent-permission findings, with optional redaction.
POST
/v1/feedbackRating and outcome signals. Never accepts prompt text.
GET
/v1/healthService status and engine version. No key needed.
GET
/v1/openapi.jsonOpenAPI 3.1 schema for code generation.
- Bearer API keys. Only hashes are stored server-side.
- Prompts and findings are never logged or stored.
- Per-key rate limits, 64 KB body limit, consistent error objects.
curl -s https://<your-intruva-api>.workers.dev/v1/optimize \
-H "Authorization: Bearer $INTRUVA_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"prompt": "Our checkout API got slow after the deploy, fix it. Don't change the response format.",
"task_type": "auto",
"mode": "quick",
"target_model": "generic",
"answers": []
}'MCP server: works with any model
Give your AI assistant a preflight step. Any client that speaks the Model Context Protocol can call Intruva to diagnose, optimize and security-scan a prompt before acting on it, whichever model it runs. Runs locally over stdio. Download it below, or let npx fetch it with the configs here.
// claude_desktop_config.json
{
"mcpServers": {
"intruva": {
"command": "npx",
"args": ["-y", "--package=https://intruva.onrender.com/downloads/intruva-mcp.tgz", "intruva-mcp"]
}
}
}Tools
intruva_analyze_promptRanked missing context, questions, assumptions, security.
intruva_optimize_promptOptimized prompt with preservation checks. Accepts answers for the clarify loop.
intruva_scan_promptSecurity preflight, with optional redaction.
intruva_list_task_typesTask blueprints and the context fields each one checks.
Typical agent loop: analyze, ask the user the questions, optimize with the answers, then act.
Download the MCP server
v0.1.0Node >=20 · runs locally · no account neededIncludes the open-source MCP SDK and zod; third-party licenses. Verify a download with shasum -a 256 intruva-mcp.tgz.
Chrome extension
Optimize right where you write: ChatGPT, Claude and Gemini. It reads the composer only when you click or use the shortcut, never sends your message for you, and always lets you revert.
1. Install
Pilot build: download the zip from the GitHub release, then load it unpacked at chrome://extensions with Developer mode on.
2. Write as usual
Draft your prompt in ChatGPT, Claude or Gemini. The Intruva button appears beside the composer.
3. Preflight
Click it or press Alt+Shift+I. Review the gaps and security findings, then replace, copy or revert.
- Runs on your device by default
- Access only to the three supported sites
- No telemetry, no prompt storage