Manufact Turns MCP From Open-Source Traction Into Cloud Infrastructure

Why it matters

Manufact shows how open-source AI infrastructure companies are turning protocol adoption into paid cloud products before the market settles on winners.

Manufact is pushing beyond its open-source mcp-use SDK with Manufact Cloud, a production platform for building and deploying Model Context Protocol (MCP) apps and servers across ChatGPT, Claude, coding agents, and internal AI tools.

Manufact says Cloud covers the jobs that begin after a local demo: MCP hosting, cross-client testing, publishing checks aligned with ChatGPT Apps Store and Claude Connectors requirements, Cloud Inspector for tracing and debugging MCP traffic, embeddable public chat surfaces, and analytics for usage, latency, and reliability. The open-source wedge is visible: the mcp-use repository shows 10.2k GitHub stars, and the homepage highlights developer usage at IBM, Intuit, NVIDIA, Oracle, Red Hat, Verizon, Elastic, 6sense, Innovaccer, and Tavily.

Language on Manufact's site is careful. Those logos are framed as developers using its tools, not a list of paying customers. The company has not disclosed revenue, deployment volume, valuation, headcount, or how many open-source users have converted to Manufact Cloud.

From SDK to a control plane

Manufact positions mcp-use as the on-ramp and Manufact Cloud as the place to ship. The company pitches a lifecycle that starts with scaffolding an app or server, continues through testing and inspection, and ends with distribution to user-facing surfaces where agents already work.

What Manufact Cloud actually ships

• Build: Scaffold with the mcp-use SDK, start from templates, install a coding-agent skill, or use the company's Vibe product to describe an app and generate a project.
• Ship: Connect a GitHub repo to enable automatic deployments on each push via the GitHub App, get per-branch preview URLs, and use custom domains with SSL.
• Observe and test: Use Cloud Inspector to trace, replay, and debug MCP traffic, fire tool calls, inspect JSON-RPC, and swap models across GPT, Claude, and Gemini, with automatic evals.
• Distribute: Run cross-client tests and publishing checks for ChatGPT Apps Store and Claude Connectors, and generate submission assets alongside an embeddable public chat surface.

The commercial question

Open-source distribution and a protocol tailwind are meaningful assets, but conversion is the test. Developers star SDKs and clone examples long before budget owners pay for hosting, observability, and workflow automation. Many target teams already have cloud infrastructure, internal platform groups, or security constraints around agent access to company systems. Manufact's bet is that MCP-specific depth, compatibility knowledge, and speed across many clients will make Cloud worth adopting.

The market is fragmented today. ChatGPT, Claude, Gemini Enterprise, Copilot 365, Codex, Claude Code, Cursor, VS Code, OpenAI Agents, Claude Agent SDK, Mastra, LangChain, Vercel AI SDK, and CopilotKit all appear in Manufact's own target-surface language. A toolchain that can test once across those clients and prepare a submission pack has an obvious use case if marketplaces and connectors become real distribution channels.