Hey there, here's what we shipped in October.
Score any MCP server on the public leaderboard
1,088 MCP servers, scored and ranked by category. Paste any URL for a free report with fixes. You can share it on socials and climb the leaderboard to be featured on the home page!
The score is broken down into these categories:
Server Metadata: how comprehensive your server information is.
Discoverability: can your server be discovered by users? Are you on major marketplaces?
Protocol, Tools, MCP Apps: is your server following the protocol and its best practices in terms of tool annotations, security and MCP Apps support?
Tool Quality: we run an evaluator on all your tools, reporting issues in descriptions, titles and argument definitions to flag vague or missing declarations.
Try to hill-climb the leaderboard, your MCP will benefit!
Submission packs for ChatGPT plugins
OpenAI moved app submissions to plugin ZIPs at DevDay. We rebuilt Submission pack for that.
Give it your server URL and it generates a compliant ZIP file that can be uploaded straight into OpenAI. This includes test cases, server metadata, example prompts and demo recording video.
Tip: we allow you to run the test cases on Codex and ChatGPT before you submit, directly on Manufact!
Generate your submission pack →
See who uses your MCP
Analytics got a Users card that shows you the live count of users that are using your MCP.
You can see important metrics such as the number of sessions and tool calls per user, and even inspect each user's sessions to debug and understand user behavior.
Also shipped
Publishing checks for each AI client →
Separate ChatGPT/Codex and Claude scores, reported to GitHub on automatic runs.
Global server region →Run your server in the US, EU, and APAC, served from the nearest region.
Actor names in Sessions →Opt in per server to show caller names and emails.
Run or stop all test suites →Start every suite at once, or cancel runs mid-flight.
Charts for alert incidents →See the metrics around each alert, before and after it fired.
Manufact for Startups →Teams that raised under $2M can apply for the Startup plan.
Over 72 million tool calls
Up from 21 million in our September issue. Thank you for building with us.
Make Something Agents Want at Y Combinator
On 17–18 October we sponsored the Best MCP Server for Agents track at the YC hackathon. Across eight tracks, 80% of the 180+ builders built an MCP with Manufact. Congrats to our winners:
1st · Thump: sensors in a football helmet that log collision impact, served to coaches through an MCP App.
2nd · Pareto: an escrow MCP App for agent-to-agent transactions.
3rd · Project Y: a shared memory layer where coding agents post what they've shipped.
Thanks to Y Combinator for hosting and to Tobin South for the fireside chat.
Customer shoutout
How Alpha Vantage automated MCP hosting and QA
Alpha Vantage (YC S18) planned to host its MCP server on AWS and hand-build test suites for every AI client. They went with Manufact instead and now run hosting, deploys, testing and analytics in one place.
From the blog
OpenAI MCP Extensions: bring your app into ChatGPT as a plugin
We met the MCP Extensions team at OpenAI DevDay and wrote up how they work: your plugin can open from the ChatGPT sidebar, or work on files in your own editor. mcp-use v2.8.0 supports Plugin Extensions fully.
Why does my MCP server work in Inspector but fail in ChatGPT or Claude? →
Inspector passes, ChatGPT fails. We go step by step from connection and auth to tools and UI.
How to generate an MCP App submission pack for ChatGPT →
What's in the pack and what we check before submitting.
Meme of the month
What should we build next?
Email us at [email protected] if you have any questions!
Pietro
Manufact
