Ship and test your MCP plugin on ChatGPT, Claude, Cursor, Grok and MuseChatGPTClaudeCursorGrokMuseChatGPT

Manufact is the platform to ship and test MCP plugins. Deploy from GitHub or the CLI, run publishing checks and generate a submission pack for the ChatGPT, Claude and other plugin marketplaces, monitor usage with MCP-first analytics, and run evals against every harness and client. It's built by the team behind the open-source mcp-use SDKs for TypeScript and Python and the MCP Inspector.

mcp-use is an open-source framework for building MCP servers and MCP Apps, made by the same team. Manufact is the platform, and it doesn't require mcp-use: you can deploy and test MCP servers built with any framework, in any language, on Manufact.

Under 60 seconds from git push to a live MCP endpoint. Connect the GitHub App once and every push deploys automatically, or use npx mcp-use deploy from the CLI. An agent can also drive deployments through Manufact's official MCP server.

Any language and any MCP framework. TypeScript and Python servers get build presets, including mcp-use and FastMCP, and a Dockerfile option covers everything else.

Yes. Every branch gets its own MCP URL (<slug>--br-<branch>.run.mcp-use.com/mcp), so reviewers can point a real client at the change before merge. Available on every plan.

Manufact runs six categories of publishing checks mapped to marketplace requirements: protocol and discovery, tool conformance, security and policy, metadata and configuration, domain/TLS/CSP, and assets. Each failed check comes with fix guidance and an autofix flow. E2E checks then run your server live inside ChatGPT and Claude to verify tool calls and widget rendering, and a generated submission pack produces listing copy, tool justifications, and reviewer test cases.

An eval is a test suite for your MCP server: a representative task, an optional rubric, and a passing threshold (70% by default). Manufact runs the task in real clients and coding-agent harnesses across the models you choose, and an LLM judge scores each run from the recorded evidence: the tool calls and arguments, the agent's replies and, in ChatGPT and Claude, a screenshot of the rendered MCP App.

ChatGPT, Claude and the Inspector run in a browser and also check MCP Apps UI. Claude Code, Codex CLI and Pi test tool use, each in a fresh sandbox with the real CLI and the models you select, including GPT, Claude and Gemini models. Teams with access to native client apps can also run plugin lifecycle targets, such as Codex on Windows or ChatGPT on Android, which install your plugin and exercise the full flow.

Yes. Suites can gate CI, so a regression in any client or model blocks the merge, and you can run every suite on demand before a release. Each run comes with a report of the problems attributed to your server and a fix prompt to paste into your coding agent, then you re-run to confirm the fix.

Yes. MCP evals are available on every current plan, including Free. Runs start at $0.70 each, with graduated volume discounts, and draw from your organization's shared usage allowance.

Tool calls, sessions, error rates with issue triage, and p50/p95/p99 latency per tool, resource, and prompt, broken down by client (ChatGPT, Claude, Cursor and more), version, and country. You also see the real users behind each session, AI-discovered goals that show what they are trying to get done, and feedback that users and agents send from the chat through a send_feedback tool.

Yes, on the Enterprise plan. Book a call and we'll connect your existing remote MCP server so you get evals, publishing checks, and the submission pack against it. Hosting on Manufact adds deploys from GitHub, preview environments, and gateway-level analytics.

The Free plan is $0 and includes $5/month in usage. Pay as you go has no subscription fee and includes $10/month in usage, with additional usage billed within your chosen budget. Pro is $300/month or $250/month billed $3,000 annually, with $300/month in included usage. Enterprise terms are agreed individually. Usage rates are graduated; see the pricing page for current rates. Discovery requests such as tools/list are not billed as tool calls.