Corsair – open source integration layer for AI agents, built on MCP

If you're building an AI product that needs to talk to Gmail, Slack, Notion, Stripe, whatever, the actual agent logic is the easy part. The hard part is everything underneath it: OAuth flows that are different for every provider, token refresh, webhook signature verification, multi tenant credential isolation, and APIs that quietly change and break your integration six months later. Teams either burn months building and maintaining that plumbing themselves, or hand it to a closed source platform and get stuck waiting on someone else's roadmap when they need an integration that isn't already built.

What we built

Corsair is an open source, TypeScript native integration layer built on MCP. You can self host the SDK for free on your own infrastructure, or use our hosted Hub if you'd rather not run it yourself. Either way, none of your credentials or user data sit with us. Corsair resolves credentials internally at call time, your agent only ever sees method names and results, and destructive actions require explicit approval before they execute so your users get a review link instead of a surprise.

Because it's open source and Apache 2.0 licensed, if we're missing an integration you need, you're not stuck filing a feature request and waiting. Fork it, open a PR, or scaffold a new plugin yourself.

Where we're at

We went through YC (W25), we're at 9.9k stars on GitHub, and the free Hobby tier has no credit card requirement, unlimited tool calls, 50 connections, and up to 3 team members.

Repo : https://github.com/corsairdev/corsair

Website: https://corsair.dev/

Would love feedback from anyone here who's building agent products and has hit this same integration headache. What's been your workaround so far, roll your own, or use one of the closed platforms?