Manufact

Manufact

Manufact (formerly mcp-use) is the MCP cloud for deploying MCP servers and apps from GitHub, testing them with evals inside ChatGPT, Claude, Claude Code and Codex, preparing marketplace submissions, and tracking per-tool analytics. Free tier; usage-based Pay as you go; Pro $300/month.

0.0 (0)
Visit Website
Founder
Manufact, Inc.
Active
PlanFree - $0 - $5/month included usage, 1 project, 1 member, 30-day analytics, hard limits · Pay as you go - no subscription - $10/month included usage, 10 projects, spending limits; tool calls $0.10 per 1K (graduated), evals $0.70/run · Pro - $300/month or $250/month billed annually - $300/month included usage, unlimited projects, 90-day analytics, cold-start prevention, all regions, SOC 2/ISO 27001
BusinessB2B

About Manufact

What Manufact is

Manufact is a cloud platform for shipping MCP servers and MCP apps: it hosts them, tests them inside the AI clients people actually use, prepares them for the plugin marketplaces of ChatGPT, Claude, Cursor, Grok, Muse and Gemini, and shows how they are used once live. MCP, the Model Context Protocol, is the open standard that lets an AI assistant call a product's tools; Manufact's argument is that an MCP server is becoming a product surface in its own right, like a website or a mobile app, and needs the same deploy, test, release and analytics loop.

The company was previously known as mcp-use, after its open-source framework, which it still maintains. Manufact, Inc. was founded in 2025 by Pietro Zullo and Luigi Pederzani and later joined by CTO Enrico Toniato, a former IBM Research tech lead, according to its February 2026 funding announcement. It went through Y Combinator (S25) and raised a $6.3M seed round led by Peak XV, with Liquid 2 Ventures, Ritual Capital, Pioneer Fund, Y Combinator and angels including Supabase's co-founder and COO. Its open roles are based in San Francisco. The product launched on Product Hunt on November 19, 2026 under the tagline "Ship and test your MCP to every plugin marketplace".

What the platform does

  • Hosting. Import a GitHub repository (or start from a Starter, MCP Apps or Blank template) and Manufact builds it and deploys it to a live MCP URL, with preview deployments per branch, custom domains and, on Pro, cold-start prevention and US, EU and APAC regions.
  • Cross-client evals. You write a task ("create a high-priority issue in the Mobile project and give me the link") and pass criteria; Manufact runs it in ChatGPT, Claude and its Inspector in a browser, and in Claude Code, Codex CLI and Pi in fresh sandboxes with the models you pick, then an LLM judge scores each run from the tool calls, replies and, for MCP Apps, a screenshot of the rendered widget. Suites can gate CI.
  • Publishing checks and submission pack. Audits a server against ChatGPT Apps store and Claude connector requirements, and generates the assets a submission needs: a demo video, test cases, metadata and icon.
  • Analytics and Cloud Inspector. Tool calls, sessions, error triage and p50/p95/p99 latency per tool, broken down by client, version and country, plus traces you can replay and AI-discovered user goals.
  • Open-source mcp-use. The MIT-licensed mcp-use framework (about 11,000 GitHub stars when we checked) provides a TypeScript and Python SDK for MCP servers, React views for MCP Apps, a browser Inspector, a client CLI and tunnelling. Manufact is where it deploys by default, but the framework runs anywhere Node runs.

We built and called an MCP server with mcp-use

On November 20, 2026 we installed the framework's current release, [email protected] (published the day before), in a clean Node 22 sandbox and wrote a one-tool server, without the scaffolder and without a Manufact account:

import { MCPServer } from "mcp-use";
import { z } from "zod";
const server = new MCPServer({ name: "tlh-demo", title: "TLH demo", version: "1.0.0" });
server.tool({ name: "word-count", description: "Count words in a text",
  inputSchema: z.object({ text: z.string() }),
  outputSchema: z.object({ words: z.number() }),
  annotations: { readOnlyHint: true, destructiveHint: false, openWorldHint: false } },
  async ({ text }) => { const words = text.trim().split(/\s+/).filter(Boolean).length;
    return { content: [{ type: "text", text: `${words} words` }], structuredContent: { words } }; });
await server.listen(3011);

Then we spoke raw JSON-RPC to it with curl, the way any MCP client does over Streamable HTTP:

$ curl -s -X POST localhost:3011/mcp -H 'Content-Type: application/json' \
    -H 'Accept: application/json,text/event-stream' \
    -d '{"jsonrpc":"2.0","id":3,"method":"tools/call",
         "params":{"name":"word-count","arguments":{"text":"ship and test your MCP"}}}'
event: message
data: {"result":{"content":[{"type":"text","text":"5 words"}],"structuredContent":{"words":5}},"jsonrpc":"2.0","id":3}

# server log
initialize curl/1 /mcp 200 client=curl/1 15ms
tools/list /mcp 200 client=unknown 6ms
tools/call word-count /mcp 200 client=unknown 3ms

What we took from it:

  • Zod in, JSON Schema out. The tools/list response carried both an inputSchema and an outputSchema in JSON Schema 2020-12, generated from the Zod objects, plus the read-only and destructive hints that ChatGPT and Claude use when deciding whether to ask the user before a call.
  • Protocol 2025-06-18, tools, prompts and resources. The initialize reply advertised protocol version 2025-06-18 with tools, prompts, resources (with subscriptions) and logging, and answered as an SSE stream.
  • A landing page for free. A plain browser GET on /mcp returned an HTML page titled "TLH demo MCP Server — Connect with Claude Code, Cursor & VS Code" with connection instructions.
  • Heavier than it looks. Installing mcp-use and zod pulled in 45 top-level packages and 87 MB of node_modules, because the package also carries the Inspector, React views and OAuth helpers.

Pricing (checked November 20, 2026)

PlanFeeIncluded usageNotes
Free$0$5/month1 project, 1 member, 30-day analytics, hard limits, no card
Pay as you goNo subscription$10/month10 projects, 10 members, spending limits
Pro$300/month, or $250/month billed $3,000 a year$300/monthUnlimited projects and members, 90-day analytics, cold-start prevention, all regions, SOC 2 and ISO 27001, Slack support
EnterpriseCustomContractedSSO/SAML, RBAC, self-hosting, custom SLA, connecting servers hosted elsewhere

Usage rates: tool calls $0.10 per 1,000 up to 1M a month, then $0.09 (1M–10M), $0.08 (10M–50M) and $0.07 beyond; evals $0.70 a run ($0.63 after 1,000, $0.56 after 10,000); publishing checklist $0.90 a run; submission pack $0.50; build minutes $0.07; egress $0.15/GB; LLM tokens $5 per million. Discovery requests such as tools/list are not billed as tool calls.

Worked example: a month for a mid-sized MCP app

Take a server handling 2 million tool calls a month, with an eval suite of 10 tasks run 4 times in each of 4 clients before each of 2 releases:

Tool calls  first 1M x $0.10/1K = $100
            next  1M x $0.09/1K = $90        -> $190
Evals       10 tasks x 4 runs x 4 clients x 2 releases = 320 runs x $0.70 = $224
Usage total                                         = $414

Pay as you go: $414 - $10 included                  = $404/month
Pro (annual):  $250 fee, $300 included -> $250 + ($414 - $300) = $364/month

At this volume Pro on annual billing is already cheaper than Pay as you go, before counting its cold-start prevention and regions. Below roughly $250 a month of usage, Pay as you go wins. Bandwidth, builds and LLM tokens for evals are left out because they depend on the server; the eval count is our assumption, not a vendor figure.

Manufact vs the alternatives

Manufactmcp-use self-hostedVercelComposio
What it isMCP-specific hosting, evals, marketplace prep and analyticsThe same open-source framework on your own infrastructureGeneral frontend and serverless hostingManaged integrations and tools for AI agents
Who writes the serverYou (any framework; templates for mcp-use)YouYouComposio, for its catalogue of third-party apps
Cross-client evalsYes: ChatGPT, Claude, Claude Code, Codex CLI, PiNo (Inspector and client CLI for manual checks)NoNot its focus
Marketplace submission helpPublishing checks and submission packNoNoNo
MCP-level analyticsPer tool, client and user sessionWhatever you buildHTTP-levelTool execution logs

Who it's for

SaaS companies turning their product into a ChatGPT app or a Claude connector and wanting it accepted on the first submission; teams already building with mcp-use that want a production path; and platform teams running several internal MCP servers that need previews, analytics and regression evals across clients. Its customer stories include AgentMail and French and Spanish mortgage brokers Meilleurtaux and iAhorro.

Limits and trade-offs

  • Bring-your-own-hosting is Enterprise-only. Evals, publishing checks and the submission pack against a server hosted elsewhere require the Enterprise plan; on other plans the server must be deployed on Manufact.
  • Evals add up quickly. At $0.70 a run, a matrix of tasks x clients x repetitions grows multiplicatively, and LLM tokens are billed on top at $5 per million.
  • Compliance sits on Pro. SOC 2 and ISO 27001, cold-start prevention and non-default regions are Pro features; SSO, RBAC and self-hosting are Enterprise.
  • Marketplace rules move. Publishing checks track ChatGPT and Claude store requirements, which change; passing the checklist is preparation, not a guarantee of approval.
  • Fast-moving framework. mcp-use shipped two releases on November 19, 2026 alone; pin versions in production.

Related on TechLogHub

Value & Audience

Value Proposition

One platform to deploy an MCP server, prove it works in every major AI client, get it ready for the ChatGPT and Claude marketplaces, and see how people use it once it is live.

Problem Solved

MCP servers behave differently across ChatGPT, Claude and coding agents, marketplace submissions get rejected for missing requirements, and teams have no MCP-level analytics once a server is live.

Share this product

Product founder?

Get Badge

Tech Stack

Social & Links

Loading ratings...

Loading comments...

Gallery

Manufact screenshot 1

Feedback & Roadmap

Loading feedback…

Manufact FAQ

Common questions about Manufact's features, pricing, and use cases

Is Manufact free to use?

Manufact has a free tier plus paid plans, starting at Free - $0 - $5/month included usage, 1 project, 1 member, 30-day analytics, hard limits. Plans listed: Free - $0 - $5/month included usage, 1 project, 1 member, 30-day analytics, hard limits; Pay as you go - no subscription - $10/month included usage, 10 projects, spending limits; tool calls $0.10 per 1K (graduated), evals $0.70/run; Pro - $300/month or $250/month billed annually - $300/month included usage, unlimited projects, 90-day analytics, cold-start prevention, all regions, SOC 2/ISO 27001.

What are some alternatives to Manufact?

Popular Developer Tools alternatives to Manufact include CodeSpy, DevOp and Global TTFB Checker.

Manufact appears in

Newsletter

One email a week

TechLogHub tracks 914 tools and 611 open-source projects. Get the week's new launches, release roundups and open-source picks — nothing else.

Get the TechLogHub digest

One email a week: new and trending developer tools, fresh comparisons, and what shipped. Unsubscribe in one click.