Integration lifecycle infrastructure for AI agents

Give your agents one place to discover integrations, build and test workflows, and promote them safely from development to production, all without leaving the chat.

Connect your agent View GitHub

⌘Agent

velane · MCP

Open source · AGPL-3.0800+ integrationsBun & Python runtimesIsolated sandbox per invocation3 environments: dev · staging · prod

Integrations

800+ tools. Zero tokens in your code.

Your agent discovers connected accounts and calls any API through Velane — no OAuth flows, no credentials, no glue code.

Salesforce

Zendesk

HubSpot

Stripe

Notion

Slack

GitHub

Linear

Browse the full catalog on Nango

Who it's for

Built for developers who ship with AI

Whether you're a solo developer or part of a larger team, Velane fits the way you already work.

Solo developers

You're building an AI agent and don't want to maintain a backend just to call Stripe or Slack. Velane handles the runtime, OAuth, and deployment pipeline so you can stay in your editor.

Ship integrations in one agent conversation
No OAuth plumbing or credential management
Open source — self-host or use the hosted version
Sandboxed execution, no infra to manage

Teams

Your agents need shared credentials, audit trails, and controlled rollouts. Velane gives every team member's agent the same integrations without anyone touching a secret.

Shared credential store — one OAuth connection per provider
Environment promotion: dev → staging → prod
Invocation logs and audit trail per tenant
Role-based access: invoke, manage, admin scopes

How it works

Your IDE is now your deployment pipeline.

One MCP connection. Your agent handles the rest — writing, testing, and shipping production workflows autonomously.

Your Agent

Cursor · Claude Code

Velane MCP

mcp.velane.sh

Sandbox

Bun · Python

Integration

800+ APIs

Create & Connect

Your agent calls list_connections to discover your OAuth accounts, fetches live API docs for any provider, then writes Bun or Python code and saves it as a versioned workflow.

Run & Debug

The agent invokes the workflow on dev, reads execution logs with get_logs, and autonomously fixes errors — all without leaving the chat window. Each run is isolated in an ephemeral sandbox.

Promote

Once tests pass, the agent calls publish_snippet with env: "prod". Your workflow goes live at a stable HTTP endpoint — versioned, audited, and rollback-ready.

Setup

Two lines to connect. One handler to ship.

Add Velane to your MCP config, then your agent writes real Bun or Python code. No credentials in code — the proxy injects auth automatically.

mcp.json

{
  "mcpServers": {
    "velane": {
      "url": "https://mcp.velane.sh/mcp",
      "headers": {
        "Authorization": "Bearer vl_xxxx"
      }
    }
  }
}

stripe-to-hubspot.ts · agent-generated

import { integration } from '@velane/integrations'

export default async function handler({ customerId }) {
  const stripe  = integration('stripe')
  const hubspot = integration('hubspot')

  const customer = await stripe.get(`/v1/customers/${customerId}`)

  await hubspot.post('/crm/v3/objects/contacts', {
    properties: { email: customer.email },
  })

  return { synced: true }
}

800+

Integrations via Nango

Runtimes: Bun & Python

Environments: dev · staging · prod

AGPL

Open source licensed

Comparison

Why not just use...

Lambda handles execution. Pipedream handles workflows. Neither was designed for AI agents that write, test, and ship code autonomously.

Feature	DIY / Lambda	Pipedream	Velane
Agent-native via MCP
Zero OAuth setup		Partial
Isolated sandbox per invocation
Execution logs in your agent
Environment promotion (dev → prod)
Write code in your editor
Self-hostable