Corsair
Getting Started

Introduction

The Integration Layer for AI Agents

Corsair gives your agent safe access to hundreds of integrations. It handles all of the plumbing — auth, data storage, webhooks, token refresh, rate limiting — so the only code you write is the part specific to your use case.

corsair.ts
import { createCorsair, slack, github, gmail, linear, googlecalendar } from 'corsair';

export const corsair = createCorsair({
    plugins: [slack(), github(), gmail(), linear(), googlecalendar()],
    database: db,
    kek: process.env.CORSAIR_KEK!,
});

Connect it to your agent and start prompting:

Invite Jim to next Thursday's sales call. Tell him over Slack too so he
can accept it. Let me know when he does.

Corsair exposes four MCP tools to your agent — no matter how many plugins you have:

  • corsair_setup — check auth and get credential instructions
  • list_operations — discover every available API endpoint
  • get_schema — inspect the parameters for a specific endpoint
  • run_script — execute a script with corsair in scope

What your agent can do

Once connected, your agent can reason across all your integrations at once:

"Summarize my unread emails from customers, open a Linear issue for anything that looks urgent, and post the digest to #standup."

"When a new GitHub issue is labeled bug, create a matching Linear ticket and notify the on-call engineer in Slack."

The second one is a live webhook workflow. Corsair handles the event routing and your agent handles the logic.

Why Corsair?

Integrations are 95% plumbing

If you look at any codebase with integrations, 95% of the code is identical across every project. OAuth flows, token refresh, webhook verification, data normalization — it's all the same. Corsair handles the 95% so you write only the part that's actually unique to your use case.

Your data, your database

Everything stays in your own database. No data leaving to a third-party service, no vendor lock-in, no per-seat pricing on your users' data.

Built for agents

Corsair is designed around how agents actually work — discover, inspect, execute. The four MCP tools work the same way regardless of which framework you use.

Scales to multi-tenant SaaS

The same setup works for building products. Set multiTenancy: true, call corsair.withTenant(teamId), and every API call is automatically scoped to the right credentials.

const tenant = corsair.withTenant("team_abc");
await tenant.slack.api.messages.post({ channel: "C01234567", text: "Hello!" });

Supported integrations

Slack · Linear · HubSpot · Gmail · Google Calendar · GitHub · Google Drive · Google Sheets · PostHog · Resend · and more

Supported frameworks

Claude Agent SDK · Anthropic SDK · OpenAI Agents · Vercel AI SDK · Mastra · OpenAI · and more

Get started

  • Installation — Install, migrate, and run your first prompt
  • MCP Adapters — Connect your agent framework
  • Plugins — Available integrations and their endpoints