AutoMCP

Automatically generate MCP servers from your APIs. Turn backend APIs into agent-ready MCP servers.

Comment

AutoMCP: Automating MCP Server Generation with AI

The Inspiration 💡

Creating MCP servers manually is tedious—writing boilerplate code, defining tool schemas, handling authentication. We asked: what if we could automate this?

AutoMCP bridges API documentation and AI agent integration. Developers upload OpenAPI/Swagger docs and get a production-ready MCP server in minutes, not hours.

What We Learned 📚

Prompt Engineering: Structured, context-rich prompts with clear examples produce better code. We learned to break complex tasks into focused prompts (types → tools → server).

Schema Resolution: OpenAPI's $ref references create complex dependency graphs. We implemented recursive resolution with cycle detection to handle nested schemas.

Error Resilience: External APIs fail. We added retry logic, model validation, and auto-detection to gracefully handle Gemini API issues.

Type Safety: Generating TypeScript that compiles requires careful schema-to-type conversion and handling optional vs required parameters.

How We Built It 🏗️

Architecture

Upload → Parser → Transformer → Gemini AI → Generator → ZIP Package

1. Parsing Layer (src/parsers/)

  • OpenAPI 3.x parser (JSON/YAML)
  • Extracts endpoints, parameters, schemas, auth schemes
  • Recursive $ref resolution with cycle detection

2. Transformation Layer (src/transformers/)

  • Normalizes data for AI consumption
  • Reduces token usage while preserving context
  • Creates endpoint summaries

3. Gemini Integration (src/gemini/)

  • Multi-stage code generation:
    • Types → Tool Definitions → Implementations → Server
  • Retry logic with exponential backoff
  • Model validation and auto-detection
  • Extracts code from markdown responses

4. Package Generator (src/generators/)

  • Assembles complete npm package
  • Generates package.json, tsconfig.json, config files
  • Creates ZIP archive for distribution

5. REST API (src/api/)

  • Express.js endpoints: upload, generate, download, status
  • Local file storage with UUID sessions
  • File upload handling with multer

Tech Stack

  • TypeScript + Express.js
  • Google Gemini AI for code generation
  • js-yaml for OpenAPI parsing
  • @modelcontextprotocol/sdk for MCP protocol

Challenges We Faced 🚧

1. Model Availability: gemini-3.0-pro didn't exist, gemini-1.5-pro wasn't available. Solution: Implemented model validation with fallback to gemini-pro.

2. URL Construction: Generated code used new URL() which failed with relative paths. Solution: Switched to string concatenation with trailing slash handling.

3. Schema References: Complex $ref graphs with potential cycles. Solution: Recursive resolution with visited set for cycle detection.

4. Token Limits: Large APIs exceeded limits. Solution: Multi-stage generation (types separately, tools individually).

5. Type Safety: Ensuring generated TypeScript compiles. Solution: Explicit schema-to-type conversion with proper annotations.

Impact 🌟

  • Time Savings: Hours → minutes for MCP server creation
  • Zero Boilerplate: Eliminates repetitive code writing
  • Protocol Compliance: Ensures MCP standards automatically
  • Production-Ready: Generated code includes error handling, types, docs

Future Enhancements 🚀

  • Postman collection support
  • Web UI dashboard
  • Batch generation
  • Cloud storage integration
  • CI/CD templates

Technical Highlights 🔬

Performance: Parallel tool generation, schema caching, lazy loading

Security: Path validation, file size limits, UUID sessions, input sanitization

Code Quality: TypeScript strict mode, comprehensive error handling, detailed logging

Built in 2 hours with passion, powered by Gemini AI.

Built With

Share this project:

Updates