Metorial wants to be the invisible plumbing under every AI agent

The demo always looks like magic. An AI agent reads your calendar, files a ticket, updates a spreadsheet, pings a colleague on Slack. What the demo hides is the week of work behind each of those verbs - a Dockerfile here, an OAuth dance there, a token that expires at the worst possible moment, and no record of which agent did what to whom. Metorial, a company in Y Combinator's Fall 2025 batch, was built for the part nobody films.

Its founders describe the product plainly: agentic infrastructure to power AI-native companies. In practice, that means Metorial sits between an AI agent and every system it wants to touch - and handles the connection so a developer doesn't have to. The shorthand the team uses is four words long: "Vercel for MCP."

MCP is the Model Context Protocol, an open standard for wiring AI models to external tools and data. It caught on fast. The trouble, as anyone who has tried to run one in production knows, is that a protocol is a promise, not a service. Someone still has to host the servers, refresh the tokens, lock down the permissions, and keep the logs. That someone, increasingly, is Metorial.

The problem it actually solves

Ask most teams why their agent isn't in production yet and the answer is rarely "the model isn't good enough." It's the plumbing. Deploying MCP servers by hand means Docker configuration, OAuth flows for every service, debugging that eats afternoons, and manual API glue. That work pushes launches out by weeks or months. Metorial's pitch collapses it into a few clicks and a single SDK call.

"Developers want to ship agents now, not build infrastructure for months."— Metorial, Launch YC

The engineering choice underneath is worth pausing on. Metorial runs MCP servers serverlessly. Each server hibernates when idle and wakes in under a second when a request arrives. You pay per request instead of paying to keep a connection warm all day. It is the same insight that made serverless functions popular a decade ago, applied to a brand-new class of workload.

Agent

Claude, Cursor, ChatGPT, Copilot

→

Metorial

Auth · Permissions · Hibernation · Tracing

→

Systems

Slack, Salesforce, Stripe, Notion, 1,200+ more

What you can actually do with it

The catalog is the front door. Metorial lists more than 1,200 verified integrations across SaaS tools, enterprise systems, data sources, and custom MCP servers. Magic MCP turns all of it into a single URL: point any MCP-compatible client at it, and the agent can reach 600+ integrations with zero configuration. For teams building their own agents, JavaScript/TypeScript and Python SDKs reduce an integration to a single function.

Then come the parts that make a security team say yes. Protoguard inspects agent requests and blocks unsafe ones - including prompt-injection attempts - before they reach anything real. Access control maps agents onto the identity systems companies already run: RBAC, SAML SSO, IAM, and scoped permissions per agent, workflow, team, or environment. And every session and tool call lands in an audit log, so there is always an answer to the uncomfortable question of which agent acted with whose credentials.

Where the effort goes without Metorial

OAuth & tokens

weeks

Docker & hosting

days

Permissions/SSO

days

Audit & tracing

ongoing

With Metorial

3 clicks

Who is using it

The early crowd looks like the classic developer-tools shape: an open-source core that engineers try on their own, and a managed runtime companies pay for once it matters. The project has passed 3,300 stars on GitHub. The team says engineers at FAANG firms, the Big 4, and American Express are among its users, and that it is piloting with leading startups and Fortune 500s. The buyers who care most are the ones shipping customer-facing agents, where a single mis-scoped permission is a real incident rather than a demo hiccup.

Open-source identity and access layer for AI agents - connect agents to real systems with consistent auth, permissions, and observability.— Metorial, GitHub

The people, and the scar tissue

Metorial was founded in 2025 by Karim Rahme, its CEO, and Tobias Herber, its CTO. Before AI, the two ran Valicit, an event-ticketing platform out of Abu Dhabi, where they sold more than two million tickets for artists including Travis Scott, 50 Cent, and G-Eazy. Ticketing is an unforgiving teacher: traffic arrives in a spike the moment a drop goes live, and the system either holds or it embarrasses you in public. That is a useful thing to have learned before deciding to run infrastructure that thousands of agents hammer unpredictably.

Herber brings a systems-software background - distributed systems, an IEEE best-paper credit - which shows up in the product's fixation on cold starts and hibernation. The company is small, four to five people, and builds in the open. Its internal tools carry names like Starbase, an MCP debugger, and Lowerdeck, a set of shared libraries.

Where it sits in the market

MCP infrastructure has become a small, busy neighborhood. Metorial shares the block with MCP-hosting and agent-integration players such as Smithery, Composio, and Toolhouse, and brushes up against older API and OAuth aggregators like Merge and Paragon. Its sharpest claim to difference is the runtime: it presents itself as the serverless option, the one that hibernates servers and starts them in under a second, with OAuth and enterprise security folded in rather than bolted on. The quiet competitor, as always in infrastructure, is the in-house team that decides to build it themselves - the option Metorial is betting most companies will eventually regret.

The business model follows the open-core script. The core platform is open-source under Apache-2.0 and self-hostable, which earns trust and adoption. The money is in the managed serverless runtime, billed per request, with enterprise tiers for governance, SSO/SAML, and support. It is a familiar arc - free where trust is built, paid where reliability is demanded - and it fits a market that is still deciding how much of its agent stack to own.

For now, Metorial is making a specific, unglamorous bet: that the winning agent-infrastructure company won't be the one with the flashiest demo, but the one whose plumbing you never have to think about. If the last decade of developer tools is any guide, that is not a bad place to stand.