For many years, software program corporations designed their merchandise for a single sort of buyer: a human being watching a display screen. Each button, menu, and dashboard existed to translate an individual’s intention right into a machine’s motion. However a small startup primarily based in San Francisco and Zurich believes that period is ending — and that the long run belongs to corporations that construct software program not for folks, however for the unreal intelligence brokers that more and more act on their behalf.
Manufact, a three-person firm that emerged from Y Combinator’s Summer season 2025 batch, introduced in February that it raised $6.3 million in seed funding led by Peak XV, the enterprise capital agency previously often known as Sequoia Capital India and Southeast Asia, which now manages greater than $10 billion in belongings. Liquid 2 Ventures, Ritual Capital, Pioneer Fund, and Y Combinator additionally participated within the spherical, alongside angel buyers together with the co-founder and chief working officer of Supabase.
The corporate’s thesis is deceptively easy and doubtlessly monumental: as AI brokers take over extra of the work that people carry out inside software program purposes — submitting expense stories, managing buyer assist tickets, writing code, reserving journey — each software program product on earth will want a brand new sort of interface designed particularly for these brokers. Manufact is constructing the open-source instruments and cloud infrastructure to make that transition attainable.
“Software program merchandise are already being accessed by and can be accessed primarily by AI brokers, or by customers via chat interfaces,” Luigi Pederzani, co-founder and co-CEO of Manufact, stated in an interview with VentureBeat. “That’s our guess. That’s our thesis. And that’s what we’re actually rooting our firm on.”
How Anthropic’s Mannequin Context Protocol grew to become the common customary for AI brokers
To know Manufact, you first have to know the know-how it’s constructed on: the Mannequin Context Protocol, or MCP, an open customary launched by Anthropic in late 2024 that has quickly change into the dominant means for AI brokers to speak with exterior software program instruments and knowledge sources.
Earlier than MCP, connecting an AI agent to an organization’s software program required customized integration work for each single instrument — a bespoke connector for Slack, one other for Salesforce, one other for a database. It was tedious, costly, and fragile. MCP standardized this course of right into a single protocol, functioning as what CIO journal just lately referred to as “the USB-C of AI” — a common connector that lets any AI mannequin plug into any software program system via a single, constant interface.
The adoption has been explosive. In December 2025, Anthropic donated MCP to the Linux Basis’s new Agentic AI Basis, co-founded with Block and OpenAI, with assist from Google, Microsoft, Amazon Net Companies, and Cloudflare. Greater than 10,000 lively public MCP servers now function throughout the ecosystem. ChatGPT, Cursor, Google Gemini, Microsoft Copilot, and Visible Studio Code all assist the protocol. Enterprise-grade deployment infrastructure exists from AWS, Cloudflare, Google Cloud, and Microsoft Azure. An estimated 7 million downloads of MCP servers happen each month.
“Nice protocols are pretty much as good as their adoption,” Pederzani stated, drawing a comparability to the cellular revolution. “We noticed the identical transition with cellular, proper? To start with, corporations have been simply creating a reasonably easy cellular app. Who would have purchased a lodge or a flight or used a checking account from a cellular app? However as time handed, the online grew to become cellular first. What we expect is that software program merchandise can be MCP first, or chat first.”
The stakes are excessive. The worldwide AI brokers market reached $7.84 billion in 2025 and is projected to surge to $52.62 billion by 2030, in response to business analysts. The MCP Dev Summit, the most important convention devoted to the protocol, takes place April 2–3 in New York Metropolis beneath the Linux Basis’s banner, with audio system from Docker, Workato, and main cloud suppliers — and Manufact can be among the many corporations presenting.
Two Italian founders, a Zurich co-working area, and an open-source library that went viral
Manufact’s origin story reads like a case research within the energy of open-source communities to validate a startup concept earlier than a single greenback of enterprise capital is raised.
Pietro Zullo and Luigi Pederzani, each initially from Italy, met at a co-working area in Zurich — the identical area that produced Browser Use, Bloom, and different startups that went via YC in earlier batches. Zullo was finding out at ETH Zurich; Pederzani was working at Morgen, an ETH spin-off AI startup utilized by groups at Spotify, GitHub, and Linear, after main a 12-engineer crew at Accenture Switzerland. Each have been winding down earlier tasks in early 2025 when MCP launched.
“We each wrote brokers previously, and it was such a multitude to jot down the instruments, the integrations,” Zullo recalled. “When MCP got here out, it appeared like the proper match for what we have been attempting to do. However solely Cursor, Claude Code, just a few closed-source purposes allowed you to truly use the protocol. I don’t suppose I’m going to do groceries or browse the web or test my emails from Cursor — it’s like, not the best code, proper? So we wrote an open-source library to mainly do what you can do in Cursor with MCP servers, however by yourself machine, by yourself software, in your individual phrases.”
They referred to as the library mcp-use, with a slogan that resonated throughout the developer group: “Join any MCP to any LLM in six traces of code.” The repository attracted 2,000 to 2,500 GitHub stars inside weeks. At present, the SDK has surpassed 5 million downloads and 9,000 GitHub stars. Organizations together with NASA, Nvidia, and SAP use the library, and Manufact claims that 20 p.c of the US 500 have experimented with it.
“The quantity of energy which you can put in six traces of code was actually staggering,” Zullo stated. The pair utilized to Y Combinator on the day of the deadline. “We have been tremendous spontaneous as a result of we had this open-source vibe and simply loved the method. We had a lot vitality from the group that was lifting us up, and we knew it was going to be high-quality.”
Inside Manufact’s plan to change into the ‘Vercel for MCP’ — from SDK to cloud in 60 seconds
Manufact’s technique borrows straight from the playbook that turned Vercel right into a multi-billion-dollar firm by offering internet hosting and developer instruments for front-end internet purposes. The analogy is deliberate: simply as Vercel made it trivially simple to deploy a Subsequent.js app, Manufact desires to make it trivially simple to construct, check, and deploy the MCP servers and MCP apps that AI brokers have to work together with software program.
The corporate gives three core merchandise. First, the open-source mcp-use SDK, out there in each Python and TypeScript, lets builders spin up a totally useful AI agent related to MCP instruments in as few as six traces of code. It helps any massive language mannequin, together with native fashions, and has integrations with LangChain and different standard frameworks. Second, a built-in inspector and testing suite permits builders to visually debug their MCP servers in a browser, view uncooked JSON-RPC visitors, and check instrument execution in a sandbox — with out connecting to a stay AI agent. Third, the Manufact Cloud platform handles deployment, scaling, authentication, entry management, and observability, permitting groups to go from a GitHub push to a manufacturing MCP server in beneath 60 seconds.
“As software program turns into extra agentic, the onerous half isn’t the mannequin anymore — it’s all the pieces round it,” Zullo stated. “We began Manufact as a result of builders have been spending an excessive amount of time on plumbing as an alternative of constructing and transport their merchandise.”
The corporate has additionally moved aggressively into MCP apps, a more recent extension of the protocol that permits builders to render interactive consumer interface parts — React widgets, knowledge visualizations, enter types — straight inside chat purchasers like ChatGPT and Claude. Manufact’s SDK lets a developer scaffold an MCP app with a single terminal command, edit React widgets, and deploy to ChatGPT in beneath a minute. This positions the corporate on the middle of a doubtlessly huge new distribution channel: ChatGPT alone has greater than 800 million customers.
5 million downloads, zero income, and a crowded discipline of cloud giants
Each open-source firm faces the identical elementary rigidity: the group that makes the venture worthwhile is just not the identical factor as a paying buyer base. Manufact has been candid about this problem.
Pederzani stated the corporate made a deliberate choice after Y Combinator to focus completely on the open-source product and group, reasonably than speeding to monetize. “A whole lot of open-source tasks leap instantly on the monetization half and sort of betray the group,” he stated. Whereas NASA, Nvidia, and different distinguished organizations use the SDK, Pederzani acknowledged they aren’t paying prospects. Manufact’s goal is to achieve $2 million to $3 million in annual recurring income by the tip of 2026, which might place it for a Sequence A fundraise.
The aggressive panorama is crowding quick. AWS, Cloudflare, Vercel, and Docker have all launched MCP internet hosting options. However Manufact’s founders argue they sit in a complementary place relative to the mannequin suppliers. “Anthropic and OpenAI are betting that their very own chat merchandise — Claude and ChatGPT — will change into the first interfaces via which individuals entry all software program,” Pederzani stated. “If that guess performs out, we are going to serve these methods. That’s going to be huge.”
Why software program corporations with out MCP servers threat changing into “dumb databases” for AI brokers
Behind Manufact’s optimism lies a darker remark in regards to the software program business that offers their pitch urgency. Pederzani argued that corporations that fail to make their merchandise accessible to AI brokers threat being lowered to “methods of report” — dumb databases that brokers question however that now not personal the consumer expertise or the client relationship.
“Now we now have prospects that come to us and say that their prospects are selecting to undertake their product over a competitor as a result of they provide an MCP server,” Pederzani stated. “On the similar time, there’s a menace right here that might put corporations to change into simply methods of data. And that is actually one thing that plenty of corporations are fearful of.”
In late February, Manufact co-hosted what it referred to as the most important MCP apps hackathon so far at Y Combinator’s headquarters in San Francisco. The occasion drew 650 purposes and 300 builders. OpenAI, Cloudflare, and Anthropic all sponsored it. Maybe probably the most telling element: eight staff from Anthropic attended — extra folks than Manufact’s personal three-person crew. The mannequin suppliers, it seems, view Manufact as an ally reasonably than a menace.
Three staff, $6.3 million, and the ambition to seize a share of each AI instrument name on Earth
For all its momentum, Manufact faces vital headwinds. The corporate has simply three staff and has not but demonstrated a scalable income mannequin. Its most high-profile customers will not be paying prospects. The $6.3 million seed spherical offers restricted runway in an business the place infrastructure corporations usually require substantial capital to achieve profitability. And the cloud suppliers which have launched MCP internet hosting options already personal the client relationships and billing infrastructure that enterprise consumers depend on.
However when requested what success seems like in two years, each founders pointed to a single metric: the share of worldwide AI instrument calls that move via their infrastructure. “Our metric is the worldwide instrument calls or servers that run on Manufact — what number of instrument calls are passing via Manufact, made by brokers,” Pederzani stated. “Like Stripe is doing for the worldwide GDP. We’re going to win if we will get an important quantity for it.”
The Stripe analogy is formidable — Stripe processes tons of of billions of {dollars} yearly and is valued at roughly $90 billion — however it captures the scope of what Manufact’s founders imagine is at stake. If MCP turns into the common customary via which AI brokers work together with all software program, the corporate that gives the infrastructure for constructing and deploying MCP servers may occupy a place of outsized affect.
“In the long run, what issues is to make one thing brokers need,” Zullo stated, riffing on Y Combinator’s well-known dictum to “make one thing folks need.” “What we’re specializing in and what we’re constructing is to assist this transition of constructing for brokers as an alternative of constructing for people.”
