DEVELOPER TOOLS • JUNE 2026
We Built an MCP Server for Prompt Engineering
Score, scan, and optimize AI prompts directly from Cursor, Claude Desktop, or Windsurf. Our free MCP server brings STCO prompt engineering to your IDE — no context-switching required.
Why We Built This
Every day, thousands of developers write AI prompts inside their IDE. They draft system instructions for Cursor, compose chain-of-thought prompts for Claude, or tweak few-shot examples for GPT-4o — all without ever leaving their editor. But the moment they want to evaluate that prompt, they have to copy it, paste it into a web app, wait for a score, then jump back to their editor. That workflow is broken.
We built the AI Prompt Architect MCP Server to eliminate that friction. It brings our entire prompt engineering toolkit — scoring, security scanning, STCO optimization, cursor-rules generation, and template browsing — directly into any IDE that supports the Model Context Protocol. No browser tab. No copy-paste. No context-switch.
And yes, it's completely free. No API key required for the offline tools, and the online tools use the same generous free tier as our web platform.
What is MCP?
The Model Context Protocol (MCP) is an open standard — originally created by Anthropic — that lets AI-powered IDEs connect to external tools and data sources through a unified interface. Think of it as "USB-C for AI tools": one protocol, many hosts. If your IDE speaks MCP, it can use any MCP server without custom integrations.
MCP servers expose tools (functions the AI can call), resources (data the AI can read), and prompts (templates the AI can use). When you install our MCP server, your IDE's AI assistant gains five new capabilities — scoring prompts, scanning for security issues, restructuring prompts into the STCO format, generating .cursorrules files, and browsing curated prompt templates — all executed locally or through our API, depending on the tool.
5 Tools We Ship
Our MCP server ships with five purpose-built tools. Three work entirely offline (no network calls, no API key), and two connect to our scoring API for AI-powered analysis.
score_prompt Online
Send any prompt and receive a 0–100 quality score with a full STCO breakdown: System clarity, Task specificity, Context richness, and Output formatting. The response includes per-dimension scores, actionable suggestions, and a letter grade (A+ through F).
Example usage in your IDE:
Returns: Score 72/100 — System: 85, Task: 78, Context: 45, Output: 62. Suggestion: "Add explicit output format constraints and provide a code example of the expected style."
This is the same scoring engine behind our web-based Prompt Scorer, now available without leaving your editor.
scan_prompt_security Online
Detects prompt injection attacks, data leakage risks, jailbreak patterns, and PII exposure. The scanner checks for 30+ known attack vectors including role hijacking, delimiter injection, multi-language obfuscation, and recursive self-reference exploits.
Example usage:
Returns: 🚨 High risk — Injection detected: "Ignore all previous instructions" is a classic role-hijacking attack. Recommendation: Add instruction-anchoring guardrails and never echo system prompts.
optimize_prompt Offline
Takes any unstructured prompt and restructures it into the STCO format — System, Task, Context, Output. This is a deterministic transformation that runs entirely on your machine with zero network calls. It parses your intent, identifies missing sections, and produces a clean, framework-compliant prompt you can use immediately.
Before → After:
Learn more about the STCO framework in our STCO Masterclass.
generate_cursor_rules Offline
Generates a tailored .cursorrules file for any tech stack. Tell it your framework, language, and coding conventions, and it produces a production-ready rules file that Cursor (or any compatible IDE) will use to guide its AI suggestions. Covers naming conventions, import styles, testing patterns, error handling, and architecture guidelines.
Example:
Produces a comprehensive .cursorrules file with 40+ rules covering file structure, component patterns, database queries, and test conventions.
This is the same engine behind our Cursor Rules Generator web tool.
list_templates Offline
Browse 11 curated prompt templates covering the most common prompt engineering patterns: chain-of-thought, few-shot classification, structured extraction, code review, creative writing, data analysis, role-play tutoring, API documentation, debugging, summarisation, and translation. Each template follows the STCO format and includes slot placeholders you can fill in immediately.
Example:
Explore all templates in our Learning Hub.
Setup Guide
Installation takes under 60 seconds. Our MCP server runs on Node.js and installs via npx — no global packages, no build steps.
Step 1: Install
Run this command in your terminal:
That's it. The server downloads, installs, and starts automatically. You'll see a confirmation message with the server's stdio endpoint.
Step 2: Configure Your IDE
Claude Desktop
Open claude_desktop_config.json (Settings → Developer → Edit Config) and add:
Cursor
Open .cursor/mcp.json in your project root (or global config at ~/.cursor/mcp.json) and add the same server block:
Windsurf
Open Windsurf's MCP configuration (Settings → MCP Servers → Add Server) and use the same npx command as above. Windsurf uses the identical JSON format.
Step 3: Verify
Open a new chat in your IDE and ask: "List the prompt-architect tools available." You should see all five tools listed. If the server isn't connecting, check that Node.js 18+ is installed and that your IDE's MCP configuration path is correct.
What's Next
This launch is version 1.0, and we're already working on what comes next. Here's what's on the roadmap:
- Prompt Versioning: Track every iteration of a prompt with diffs, scores, and timestamps. See exactly how your prompt quality evolves over time — directly from your IDE.
- Team Sharing: Share scored prompts, templates, and cursor rules with your team through a shared workspace. One engineer optimizes a prompt; the whole team benefits.
- CI/CD Integration: Run prompt scoring and security scanning as part of your build pipeline. Fail the build if a system prompt drops below a quality threshold. For end-to-end pipeline design, see our guide on AI prompt workflow automation.
- Custom Scoring Rules: Define your own scoring criteria — enforce company-specific guidelines, compliance requirements, or domain-specific quality standards.
- VS Code Extension: For developers who prefer a native extension over MCP, we're building a dedicated VS Code extension with inline prompt scoring and quick-fix suggestions.
Want to influence the roadmap? Join our GitHub Discussions or reach out on X @aaborai.
Try It Now
Install the MCP server in under 60 seconds. No sign-up required for offline tools.
Frequently Asked Questions
Is the MCP server free?
Yes. Three of five tools work entirely offline with no API key. The two online tools use the same generous free tier as our web platform — no sign-up required.
Which IDEs support the MCP server?
Any IDE that supports the Model Context Protocol: Claude Desktop, Cursor, Windsurf, and any other MCP-compatible client. The server communicates over stdio for maximum compatibility.
Do I need an API key?
Not for the offline tools. The online tools (score_prompt, scan_prompt_security) work without a key on the free tier. Add your API key in the config for higher limits.