MCP-Bridge

OpenAPI to hosted MCP endpoint + config registry + bundles

API

Open Source

Developer Tools

Hey PH! 👋 The idea for MCP-Bridge started as a frustration I hit while building AI agents that needed to talk to APIs. Every time I wanted my agent to interact with Stripe, GitHub, or Slack, I had to manually convert OpenAPI specs into MCP config format — finding the right npm package, writing the JSON by hand, copying env var names from docs. It was tedious, error-prone, and I'd repeat it for every single API. Then if I switched from Claude Desktop to Cursor or VS Code, the config format was just different enough to break things. I thought: why can't I just paste an OpenAPI URL and get back a ready-to-use MCP config? So I built the converter. It runs entirely in the browser — your spec never touches a server. That was the MVP. But then a bigger question emerged: once you have the config, where do you put it? Every AI client needs a local JSON file or a URL it can fetch. So I built the Config URL Registry — a permanent, fetchable URL for every API's config at `mcp-bridge.pages.dev/config/{api}`. Use it as-is in Claude Desktop, Cursor, VS Code, or any MCP client. That led to hosting. Teams wanted to share a single managed endpoint instead of distributing config files. So I added Hosted Endpoints — a Cloudflare Worker that wraps any OpenAPI spec as a live MCP server. Point your agent at one URL, done. Then I realized the missing piece was context. An MCP config alone doesn't tell your AI *how* to use the API well. That's cursor rules territory — framework-specific instructions that shape how the agent writes code against each API. So I bundled them into Stacks: Stripe+Node, OpenAI+Python, Supabase+Next.js, and 5 more — each with a hosted config URL, a live endpoint, and curated cursor rules. The project grew from a 5-line script into a full platform. 1015 APIs indexed. 8 bundled stacks. Free tier that actually does something useful. And 50% off first month on everything — no coupon, just auto-applied at checkout. I'd love to hear what you think — especially if you're building AI agents that talk to APIs. What's your current workflow? What's broken about it? I want to fix the things that actually annoy you. The converter runs here: https://mcp-bridge.pages.dev/con... — try it with any OpenAPI spec, it takes 5 seconds.