How good is your MCP?

Manufact adds points from tool documentation (40%), tool definitions (30%), protocol compatibility (20%), and server metadata (10%). Unavailable required sections earn zero points. For servers that provide MCP Apps, Apps contributes 20% and the required sections share the remaining 80% in the same proportions. Servers without Apps are not penalized. The report includes checks and concrete improvements for each section. It does not certify runtime reliability or marketplace approval.

Paste your MCP HTTPS URL and select Score my MCP. Your report opens while the review runs, and tool scores and feedback appear as they become available. For OAuth servers, complete the authentication prompt when required. No Manufact account is required. The server's tool metadata, documentation feedback, and score appear in the public catalog.

New reviews score every tool returned by the server. Completed tool reviews contribute to the documentation score even if some tools could not be reviewed. Each report shows the number reviewed and the total discovered. Run a new review to retry missing tools.

The review connects to your MCP server and checks its protocol, tool definitions, and app resources. It evaluates tool metadata without calling tools or changing your server. Use Manufact Cloud to test live tool behavior.

Open a report section to see failed checks and improvements. Tool documentation reviews include feedback on titles, descriptions, inputs, annotations, and output schemas. Preview and copy a fix prompt into your coding agent, verify the suggestions against your implementation, then rerun the review. No edits are applied automatically.

An MCP publishing checklist is a pre-submission audit of a live Model Context Protocol server against the protocol, tool, metadata, and MCP Apps rules used by the ChatGPT Apps Store and Claude Cloud Connectors. Manufact's free MCP checklist connects to a public or OAuth HTTPS MCP URL and reports pass, fail, and warning results before reviewers see the server.

You validate an MCP server by pointing an MCP validator at its Streamable HTTP endpoint: complete the handshake, list tools and resources, and check names, JSON Schema, annotations, and MCP Apps UI metadata. Paste the MCP URL into this checklist to run that scan.

A ChatGPT MCP app is a public MCP endpoint ChatGPT can call as tools, with optional widgets, submitted through the ChatGPT Apps Store. Run an MCP checklist first so the handshake, tool descriptions, behavior annotations, and MCP Apps resources match what ChatGPT review expects. This page is that pre-submit scan.

Claude Cloud Connectors expect a reachable MCP server, a negotiated protocol version, unique tool names within Claude's 64-character limit, human-readable titles, descriptions that fit Claude Code's metadata size limit, and capability declarations that match discovery. The Manufact MCP checklist flags those Claude-specific limits on any public or OAuth MCP URL.

An OAuth MCP server is validated after the MCP OAuth handshake, the same publishing checks as a public URL. If the checklist shows Authenticate, complete that handshake to continue the review. No Manufact account is required.

An MCP app is an MCP server that a host such as ChatGPT or Claude can call as tools, optionally rendering an MCP Apps widget from a ui:// resource. A publishing checklist checks the tool surface and the widget envelope: MIME type, widget description, and whether linked UI resources can be read.

ChatGPT and Claude expect each MCP tool to have a unique name, a description, a human-readable title, structurally valid JSON Schema, and boolean behavior annotations for read-only, destructive, and open-world. Hidden or control characters in that metadata fail review. The MCP checklist reports each of those fields.

MCP Apps widgets should be linked from tools to ui:// resources, return supported HTML MIME types, and include a widget description. The free MCP checklist reads those UI resources without invoking tools, so envelope and metadata issues show up before ChatGPT or Claude review.

An MCP server is closer to publish-ready when it completes the MCP handshake over Streamable HTTP, advertises matching capabilities, and passes the tool, resource, and applicable MCP Apps checks used in ChatGPT and Claude listings. Manufact's MCP Quality Score adds weighted section points on a 0 to 100 scale. Unavailable required sections earn no points; MCP Apps is optional. A score is feedback on these checks, not a guarantee of marketplace approval.

An MCP Inspector is a debugger for calling tools, browsing resources, and watching JSON-RPC. An MCP checklist or MCP validator is a publishing-readiness audit against ChatGPT Apps Store and Claude Connector rules. Use Inspector while you build; use this Manufact checklist to score the same server before you submit.