Manufact's mcp-use library hit 5M downloads before a dollar of institutional funding — built on questions flooding in about production, not features.
ENTRY ANGLES
Build compliance-first MCP infrastructure (HIPAA, SOC 2, FedRAMP) for regulated industries — a 6–18 month certification moat · Create an enterprise MCP connector marketplace (Salesforce, SAP, Workday, Oracle) with production-tested connectors · Target SMEs without dedicated infrastructure engineers who need hosted MCP for their AI agents
VERTICALS
CAPABILITIES
Cloud infrastructure, Open-source community building, Compliance certification, Enterprise sales
Getting an AI agent to connect to your calendar, database, and Slack workspace is now three lines of code. Keeping it running in production is the problem nobody told you about.
Luigi Pederzani and Pietro Zullo started with the easy part. In 2025 they built mcp-use, an open-source library for prototyping MCP integrations with minimal setup. The library hit 3 million downloads in its first quarter and 5 million by early 2026. The GitHub issues that followed weren't about missing features — they were uniformly about production: how do you host this securely, manage secrets, rate-limit, and isolate tenants when real users are involved?
That question became Manufact. The company raised $6.3 million in seed funding led by Peak XV, with Liquid 2 Ventures, Ritual Capital, YCombinator, Pioneer Fund, and the co-founder of Supabase among investors. The product is cloud infrastructure for MCP servers: hosted deployment, authentication, secrets management, rate limiting, and multi-tenancy, so teams that prototyped on mcp-use don't rebuild their stack from scratch to go live.
Model Context Protocol is approximately 18 months old. The tools built on it are just now moving from prototype to production, and the infrastructure layer that will handle that production load is not yet decided.
That timing is the most interesting thing about Manufact. The company didn't arrive at the production infrastructure problem through theory — it arrived through 5 million downloads of an open-source library, with a precise view of what practitioners were actually struggling with. Most infrastructure companies build products they believe developers need. Manufact built a product developers were already asking for, in the exact words they used on GitHub.
Every developer platform that achieved broad open-source adoption eventually split into the same two layers: a core that stays open and a production tier that charges for reliability, operability, and compliance. PostgreSQL and Supabase. Next.js and Vercel. Redis and Redis Labs. The pattern is consistent enough that the question for MCP infrastructure is not whether this happens, but who captures the layer. Manufact's advantage is that it doesn't have to convince the market that a production problem exists — its own download numbers are the evidence.
Manufact's window closes when larger platforms decide the market is worth entering. If Anthropic, Microsoft, or a hyperscaler offers hosted MCP infrastructure as part of a broader developer platform, the available surface area contracts fast. The defensibility has to come from what accumulates before that: production deployments, enterprise compliance certifications, and connector integrations that take months to build and create real switching costs.
The compliance path is more defensible than it appears. HIPAA-covered MCP infrastructure for healthcare AI agents, SOC 2 Type II for enterprise software connecting AI to internal tools, FedRAMP authorization for government AI deployments — each takes 6 to 18 months to obtain and makes customers difficult to displace once they've built workflows on certified infrastructure. A competitor who decides to pursue compliance after Manufact is already certified will trail by more than a year in the segments where certification matters most.
The connector marketplace is the second moat. A library of pre-built, production-tested MCP connectors for enterprise systems — Salesforce, SAP, Workday, Oracle — is valuable in proportion to how good and well-maintained each connector is, not how many exist. A team building an HR automation agent doesn't want to write and maintain the Workday MCP connector themselves. The marketplace that already has it, vetted and hosted, is a compounding distribution channel: each new connector brings in teams who then deploy on the hosted platform.