Skip to Content
View all events

Build Your First MCP Connector for Muse, Meta's AI Agent

Published by O'Reilly Media, Inc.

Give agents what they need and get it listed where people can find it

  • Learn which AI assistants accept MCP connectors and apps, what each one calls them, and how their users find them
  • Understand what Muse needs from a connector, by MCP or raw API, and what Meta's guidelines reward and reject
  • Build and test a simple MCP server with an AI coding tool, a requirements doc and a prompt, and use golden prompts you collect to run evals
  • Prepare and submit a connector to Muse's directory, with a checklist you can reuse for Claude and ChatGPT

AI assistants are turning into app stores. ChatGPT, Claude, Cursor and now Meta's Muse all let people add connectors (also called apps, plugins or extensions) that bring a product's data and actions into the chat. When Meta opened Muse to connectors in September 2026, it received over 2,000 submissions in the first few days. Each assistant names these differently and has its own rules, review process and user experience.

In this two-hour course, Emmanuel Paraskakis maps that landscape, then goes deep on Muse: what makes it different, what Meta's guidelines ask for, and what a great connector does inside it. You'll watch him build a small MCP server from a requirements doc and a prompt, test it with free tools, and walk through Meta's submission form and dashboard. You'll leave with the starter repo, the prompts and a submission checklist, so you can build your own connector after the session.

This live event is for you because...

  • You're a software engineer, architect, tech lead or product manager whose product has an API, or a service people use to get things done
  • You want your product to show up inside AI assistants like Muse, ChatGPT and Claude
  • You want to understand MCP connectors well enough to build a prototype, or to lead the team that builds one

Prerequisites

  • A working idea of what an API is. No MCP experience needed.
  • Optional: comfort reading simple code, if you want to follow the build closely

Course Preparation

  • Optional: a Muse account at muse.ai.
  • Optional: a free Replit account, or your own AI coding tool (Claude Code, Codex, Cursor), to build alongside

Course Follow-Up

Schedule

The time frames are only estimates and may vary according to how the class is progressing.

Section 1: The agent experience map: how to make your product usable with AI (15 minutes)

  • Poll: Which AI assistants do you and your customers use?
  • Presentation: ChatGPT, Claude, Cursor and Muse: what each calls connectors, apps and plugins, and how users find and add them
  • Presentation: MCP basics: tools, data and UI, and what makes an agent call your tool
  • Q&A

Section 2: What's different about Muse (15 minutes)

  • Presentation: Muse's two paths, MCP or raw API, and what Meta asks for in each
  • Presentation: Meta's connector guidelines: finish the task inside Muse instead of sending people away; read, write and sensitive-write actions; what gets featured
  • Discussion: What one job should your product's connector do inside Muse?
  • Q&A

Section 3: Design your connector (30 minutes)

  • Presentation: From golden prompts to a requirements doc: tool names and descriptions an agent understands
  • Exercise: Write three golden prompts and three tool names/descriptions for your own product's connector
  • Q&A

Section 4: Build and test it (30 minutes)

  • Demo: Build a small MCP server from the starter repo with an AI coding agent, and publish it to a public URL
  • Demo: Test it with MCP Debugger and inside Muse, then run a quick eval with MCPJam
  • Q&A

Section 5: Submit and ship (30 minutes)

  • Presentation: Meta's submission form and developer dashboard, step by step, with what's changed lately
  • Demo: A real connector from submission to review
  • Exercise: Fill in the submission checklist for your product
  • Presentation: What to add before production: authentication, persistence, rate limits
  • Q&A

Your Instructor

Emmanuel Paraskakis

Emmanuel Paraskakis has been VP of Product three times and is now a founder at Level 250. If you've built or used APIs in the past 15 years, chances are you've used one of his products. He has led some of the most influential API tools: Apiary (550,000+ APIs), Swagger (750,000+ APIs) and Oracle's first cloud-native API gateway. Before that, he built large B2B API programs at Verisk, Moody's and Precisely. Today, he advises Fortune 500 CEOs and AI-forward startup founders on API strategy, developer experience, MCP products and AI agent readiness. He's a member of the OpenAPI Initiative and a featured speaker at apidays and other global technology conferences. He's passionate about helping others build products for AI, sharing what he's learned and helping organizations build what agents need.

linkedinXlinksearch

Skill covered

AI Agents