MCP Generator

Turn any public API into a useful MCP Server

Comment

Inspiration

There are APIs out there for almost everything. However, agents cannot natively access APIs, requiring developers to develop MCP servers for each endpoint. We set our target to unlock the world of APIs to any AI agent.

What it does

MCP Generator makes it easy to turn any public API into a tool that AI agents can use. Just drop in any API link or description, and our system reads the documentation, understands what the API does, and instantly creates a ready-to-use deployed MCP server on-the-fly, all from your browser. Behind the scenes, it’s powered by smart agents that read and extract the important parts of the API for you, so you don’t have to touch a single line of code.

How it's built

Parsing user search query - Exa

Fetching docs from link - Browserbase

Agents - ADK, A2A, CrewAI

Deploy MCP - Fly.io

Observability - Weave

Models - Weave Inference (Deepseek), Gemini, Claude

Process

  1. User enters search query (Eg: Please give me an MCP for Pokemon API)
  2. We use Exa to fetch all the relevant documentation links
  3. We use Stagehand + Browserbase to fetch the documentation, endpoints, ratelimits, responses etc.
  4. We use ADK and Crew AI to define agentic workflows to generate workflows, analyze the documentation, generate MCP code, etc
  5. We use A2A to facilitate communication between these agentic workflows.
  6. We get a final MCP server code as output from the workflow.
  7. We dockerize this and deploy each MCP server to Fly.io! 🚀

Challenges

  • Integrating Google ADK and A2A
  • Insufficient docs
  • CrewAI Hanging Issue with Weave

Accomplishments that we're proud of

  • Making the Google ADK work.
  • Built an pipeline from natural language to deployable MCP servers.
  • Integrated A2A agents with real-time logging and automated code generation.

What we learned

A2A protocol precision requirements and debugging challenges Framework integration complexity across multiple AI systems

What's next

We loved this project and would love to take it forward in the following ways:

  1. Add user authentication
  2. Deploy the agents as microservices on Fly
  3. Refine and develop an agentic layer that that fine-tunes according to what tool suits the user best.

Built With

Share this project:

Updates