← Back to research
·8 min read·company

Nango

Open source product integration platform with 800+ supported APIs, managed auth, and a hosted MCP server for agent tool calling — YC W23, $7.5M seed (April 2026), 10K+ GitHub stars.

Key takeaways

  • $7.5M seed led by Gradient (April 2026) on top of 10,000+ engineering teams, billions of API requests monthly, and cash-flow-positive operations — rare discipline for integration infrastructure
  • Exposes integrations as agent tools via a hosted MCP server with per-connection credential scoping, so provider secrets never enter the agent runtime
  • Code-first and self-hostable (Elastic License 2.0) with 800+ APIs and 3,000+ reusable templates, positioned as the open alternative to closed unified-API vendors

FAQ

What is Nango?

An open source platform for building product integrations — managed OAuth/auth, syncs, webhooks, and proxy across 800+ APIs, with integrations exposed to AI agents as tools via MCP.

How much does Nango cost?

Free tier for testing, Starter from $50/month, Growth from $500/month, and custom Enterprise plans, with usage-based metering on connections, proxy requests, function runs, and storage.

How does Nango support AI agents and MCP?

Nango runs a hosted MCP server at api.nango.dev/mcp that exposes action functions as tools, scoped per user connection, and works with OpenAI, Anthropic, Vercel AI SDK, LangChain, and Mastra.

How is Nango different from Composio?

Nango is a code-first product-integration platform (auth, syncs, webhooks) that added agent tool calling; Composio is agent-native first, leading with a large toolkit catalog and sandboxed execution.

Executive Summary

Nango is an open source platform for building product integrations: managed OAuth and credential handling, data syncs, webhooks, and an API proxy across 800+ supported APIs, with 3,000+ reusable integration templates. [1] Originally pitched as "a single API for all your integrations" out of Y Combinator's W23 batch, it has repositioned around AI — integrations are described in plain English and generated by coding agents, then exposed to AI agents as tools through a hosted MCP server. [2] [1]

The business is unusually healthy for integration infrastructure: 10,000+ engineering teams, billions of API requests processed monthly, several million in ARR, and cash-flow-positive before raising a $7.5M seed led by Gradient in April 2026, with customers including Replit, Mercor, and Ramp. [3] The core platform is self-hostable under the Elastic License 2.0 with 10K+ GitHub stars as of June 2026. [4]

AttributeValue
CompanyNango (NangoHQ)
Founded2022
BatchY Combinator W23
Funding$7.5M seed (April 2026, led by Gradient)
GitHub Stars10K+ (June 2026)
LicenseElastic License 2.0 (ELv2)

Product Overview

Nango sits between a product (or agent) and third-party APIs. Developers configure providers, ship a hosted "Connect" auth flow to end users, and then read/write external data through Nango's syncs, actions, webhooks, and proxy — with credentials, token refresh, retries, rate limits, and logging handled by the platform. [1] The same action functions that power product integrations double as agent tools: any LLM SDK or MCP client can call them on demand for a specific user connection. [5]

The 2026 repositioning is "build integrations with AI" — describe an integration in plain English and Nango's AI tooling generates the integration code against its 800+ providers, claiming 99.9% uptime and over 1M integrations running in production. [1]

Key Capabilities

CapabilityDescription
Managed AuthOAuth, API keys, and token refresh across 800+ APIs, with a drop-in Connect UI for end users
Syncs & WebhooksContinuous data ingestion from external APIs plus real-time event triggers
API ProxyAuthenticated passthrough requests to any supported provider
MCP ServerHosted MCP endpoint exposing integrations as agent tools, scoped per connection
AI CodegenIntegrations described in plain English, generated from 3,000+ reusable templates

Product Surfaces

SurfaceDescriptionAvailability
Cloud platformHosted dashboard, auth, syncs, webhooks, proxyFree–Enterprise
MCP serverapi.nango.dev/mcp for agent tool callingAll plans
Self-hostedRun the open source platform yourselfELv2; offering as add-on at higher tiers
[1] [5] [6]

Technical Architecture

Nango is a TypeScript codebase deployed as a hosted cloud service, with self-hosting permitted under ELv2 (no offering it as a competing managed service). [4] Integration logic lives in action and sync functions that run on Nango's infrastructure, which advertises sub-100ms schedule-to-execution latency. [1]

For agents, Nango exposes a hosted MCP server at https://api.nango.dev/mcp over Streamable HTTP. Clients authenticate with the account secret key plus connection-id and provider-config-key headers, so each MCP session is scoped to a single user's connection — provider credentials stay out of the agent runtime entirely. [5] Tools are action functions; a /scripts/config endpoint supports runtime tool discovery in Nango-native or OpenAI-compatible shapes, and docs show native integration patterns for OpenAI, Anthropic, Vercel AI SDK, LangChain, and Mastra. [5]

Key Technical Details

AspectDetail
DeploymentCloud (hosted); self-hosted under ELv2
LanguageTypeScript
Agent InterfaceHosted MCP server (Streamable HTTP) + native tool definitions for major LLM SDKs
Integrations800+ APIs, 3,000+ templates
Open SourceYes (NangoHQ/nango, Elastic License 2.0)
[4] [1]

Strengths

  • Credential isolation by design — per-connection MCP scoping keeps provider secrets out of the agent runtime; the agent only ever sees a connection ID and tool name [5]
  • Proven scale — 10,000+ engineering teams, billions of monthly API requests, and production customers like Replit and Ramp [3]
  • Financial discipline — several million in ARR and cash-flow positive before raising; the $7.5M seed extends an already-working business [3]
  • Open and self-hostable — ELv2 source with 10K+ stars gives an exit hatch closed unified-API vendors lack [4]
  • One platform for product and agent integrations — the same actions power customer-facing integrations and agent tool calls, avoiding parallel integration stacks [1]

Cautions

  • Integration platform first, agent platform second — MCP support arrived with the 2026 AI repositioning; agent-native rivals lead with larger tool catalogs and execution sandboxes [3]
  • Usage-based pricing complexity — connections, proxy requests, compute time, function runs, logs, storage, and webhooks are each metered separately, making bills hard to predict [6]
  • ELv2 is not OSI open source — fine for internal use and self-hosting, but you cannot offer Nango as a managed service [4]
  • Smaller war chest — $7.5M raised versus considerably larger rounds at agent-tooling competitors; efficiency cuts both ways if the category turns into a land grab [3]
  • Code-first posture — designed for engineering teams shipping integrations, not no-code operators

Pricing & Licensing

TierPriceIncludes
Free$0/monthTesting use, 2 environments
StarterFrom $50/month20 connections ($1/extra), 200K proxy requests, 20 hours function compute, 200K function runs
GrowthFrom $500/monthHigher limits, 10 environments
EnterpriseCustomUnlimited environments, HIPAA, priority support, self-hosting add-on

Overage is metered: $0.0001 per proxy request, sync record, and webhook; $0.00001 per log. [6]

Licensing model: Source-available under Elastic License 2.0 — free to use and self-host, prohibited from being resold as a hosted service. [4]

Hidden costs: Seven separate usage meters (connections, proxy, compute, runs, logs, storage, webhooks) mean costs scale with end-user activity, not just seats. [6]


Competitive Positioning

Direct Competitors

CompetitorDifferentiation
ComposioAgent-native with 1000+ toolkits and sandboxed execution; Nango counters with product-integration depth (syncs, webhooks), self-hosting, and stricter credential isolation
Merge / unified APIsClosed, category-specific unified APIs; Nango is open, code-first, and spans 800+ providers
Zapier / workflow toolsNo-code automation; Nango is embedded infrastructure for developers
Custom buildsNango replaces months of OAuth, retry, and rate-limit plumbing

When to Choose Nango Over Alternatives

  • Choose Nango when: you ship customer-facing product integrations and want the same layer to expose them to agents via MCP
  • Choose Nango when: self-hosting or source access is a procurement requirement
  • Choose Composio when: you are agent-first and want the largest ready-made tool catalog with sandboxed execution
  • Choose custom integration when: you need deep control over one or two critical APIs

Ideal Customer Profile

Best fit:

  • B2B SaaS engineering teams shipping many customer-facing integrations who now also need agent tool access
  • AI products that need authenticated, per-user access to third-party APIs without holding credentials in the agent loop
  • Teams that value cash-efficient, open infrastructure vendors over heavily subsidized closed platforms

Poor fit:

  • No-code teams wanting drag-and-drop automation
  • Agent builders who want a maximal pre-built tool catalog with zero integration code
  • Companies intending to resell integration infrastructure as a service (ELv2 prohibits it)

Viability Assessment

DimensionAssessment
Financial HealthStrong — several million ARR, cash-flow positive pre-raise, $7.5M seed (April 2026) [3]
Market PositionEstablished in product integrations; challenger in agent/MCP tooling against better-funded rivals
Innovation PaceActive — AI codegen and MCP server shipped with 2026 repositioning; repo pushed daily [4]
Community/Ecosystem10K+ stars, 1K+ forks, 10,000+ teams using the platform [4] [3]
Long-term OutlookPositive — efficient business with a real wedge (product integrations) into the agent-tooling market

Nango enters the MCP-platform race from a position most competitors lack: an existing, profitable integrations business. The risk is the inverse — agent-native platforms moving faster on tool catalogs and agent UX while Nango balances two product surfaces.


Bottom Line

Nango is the credible open, code-first option in the MCP integration platform category — a proven product-integrations business (800+ APIs, 10,000+ teams, billions of monthly requests) that now exposes the same integrations to AI agents through a hosted MCP server with per-connection credential scoping. [1] [3] [5]

Recommended for: Engineering teams that need both customer-facing integrations and agent tool access from one self-hostable platform.

Not recommended for: Agent-first builders who want the biggest plug-and-play tool catalog (see Composio) or no-code teams.

Outlook: A disciplined, cash-efficient bet on integrations converging with agent tooling — likely durable, even if it never wins the catalog-size race.


Research by Ry Walker Research • methodology