Sharing Your MCP

This tutorial walks you through adding an MCP to the Skola marketplace so your teammates can install it with a single command.

An MCP (Model Context Protocol server) extends AI agents with new tools and data sources. Skola handles installation, config injection, and discovery — you just need to write the metadata.

Prerequisites

Your MCP is already working locally
It's distributed via PyPI, npm, or a GitHub repo
You have skola installed and have run skola init

1. Scaffold the MCP metadata

skola create mcp my-mcp

You'll be prompted for: - Description - Distribution type (pypi, npm, github, docker) - Package name

Skola creates:

my-mcp/
├── README.md        ← usage docs
└── metadata.yaml    ← registry metadata

2. Fill in `metadata.yaml`

Open metadata.yaml. The key sections to complete:

name: my-mcp
version: "1.0.0"
description: "One-line description of what this MCP does."
author:
  team: your-team-name
  contact: "#your-slack-channel"

tags:
  - relevant-tag

distribution:
  type: pypi              # pypi | npm | github | docker
  package: my-mcp-package # pip/npm package name, or repo URL for github

# Environment variables the MCP needs
mcp_config:
  env:
    API_KEY:
      description: "API key for authentication"
      required: true
      secret: true       # masks input during install
    BASE_URL:
      description: "Base URL of the service"
      required: false
      default: "https://api.example.com"

compatible_with:
  - claude
  - opencode
  - cursor

Distribution types

Type	`distribution.package` value	Example
`pypi`	PyPI package name	`sentry-mcp`
`npm`	npm package name	`@company/my-mcp`
`github`	Full repo URL	`https://github.com/org/my-mcp`
`docker`	Docker image name	`company/my-mcp:latest`

3. Write `README.md`

Document: - What tools/resources the MCP exposes - What each environment variable does - Example interactions with the AI

4. Test the install locally

Point skola install at your local directory by temporarily editing ~/.skola/config.yaml to point to a local registry, or test the metadata validation directly:

skola publish mcp/my-mcp  # runs validation first, shows errors before opening PR

5. Publish to the marketplace

skola publish mcp/my-mcp

This forks the marketplace repo, creates a branch, copies your files into mcps/my-mcp/, and opens a PR.

6. After it merges

Your MCP appears in search results immediately:

skola search my-mcp
skola install mcp/my-mcp

Skola will: 1. Install the package via pip/npm/git 2. Prompt for all mcp_config.env variables (masking secrets) 3. Write the MCP entry to ~/.claude/mcp_settings.json (or whichever tool the user chooses) 4. Instruct the user to restart their AI tool

Tips

Do	Don't
Mark secret vars with `secret: true`	Store credentials in metadata
Provide sensible defaults for optional vars	Mark everything as required
Document each tool the MCP exposes in README	Leave README as the template boilerplate
Use semantic versioning	Leave version as `"1.0.0"` forever
Test with a fresh `skola install mcp/<name>`	Assume the install path works

Author

@AgiMaulana-GDT View all contributors →