mcp-get 0.1.0

The package manager for MCP servers — install and configure in one command.

mcp-hub

The package manager for MCP servers — install and configure in one command.

pip install mcp-hub
mcp install filesystem
mcp install github
mcp install playwright

No more manually editing JSON config files.

---

What is MCP?

Model Context Protocol is an open standard that lets AI clients (Claude Desktop, Cursor, Windsurf) connect to external tools and data sources. MCP servers expose capabilities like reading files, querying databases, or browsing the web.

The problem: installing an MCP server currently means finding the package, figuring out the right args, and manually editing a JSON config file for each AI client you use.

mcp-hub fixes this.

---

Installation

pip install mcp-hub

Quick start

# See what's available
mcp search

# Install a server (auto-detects Claude Desktop, Cursor, Windsurf)
mcp install filesystem
mcp install github
mcp install postgres

# See what's installed
mcp list

# Get details on a server
mcp info playwright

# Check detected AI clients
mcp clients

---

Commands

mcp install <name>

Installs a server and writes it to all detected AI client configs.

mcp install filesystem          # prompts for allowed path
mcp install github              # shows required env vars
mcp install postgres            # prompts for connection string
mcp install cursor --client cursor   # target a specific client only
mcp install memory --yes        # skip all confirmation prompts

mcp uninstall <name>

Removes a server from all client configs.

mcp uninstall filesystem
mcp uninstall github --yes

mcp list

Shows installed servers.

mcp list          # installed only
mcp list --all    # all available servers

mcp search [query]

Searches the server registry.

mcp search              # all servers
mcp search database     # filter by keyword
mcp search browser      # browser automation servers

mcp info <name>

Shows full details: description, runtime, required env vars, install prompts.

mcp info github
mcp info playwright

mcp clients

Lists supported AI clients and whether they are detected on your system.

mcp clients

---

Supported servers (25+)

Server	Description	Auth
filesystem	Read/write files in allowed paths	No
memory	Persistent memory across conversations	No
github	GitHub repos, issues, and PRs	Token
gitlab	GitLab repos and merge requests	Token
postgres	Read-only SQL queries	No
sqlite	Query SQLite databases	No
redis	Get/set Redis keys	No
fetch	Fetch web pages as Markdown	No
brave-search	Web search via Brave API	Key
puppeteer	Browser automation	No
playwright	Full browser automation (Microsoft)	No
firecrawl	Web crawling at scale	Key
exa	Semantic web search	Key
google-drive	Search and read Google Drive files	OAuth
google-maps	Geocoding and directions	Key
slack	Read and post to Slack	Token
notion	Read and write Notion pages	Token
linear	Linear issues and projects	Key
obsidian	Read Obsidian vault notes	No
docker	Manage Docker containers	No
context7	Library docs at query time	No
sentry	Query Sentry errors	Token
stripe	Stripe payments and customers	Key
sequential-thinking	Structured reasoning	No
aws-kb	AWS Bedrock Knowledge Base RAG	Keys

---

Supported clients (9)

Client	Config format	Config location
Claude Desktop	JSON	%APPDATA%\Claude\claude_desktop_config.json
Cursor	JSON	~/.cursor/mcp.json
Windsurf	JSON	~/.codeium/windsurf/mcp_config.json
VS Code	JSON	%APPDATA%\Code\User\mcp.json
Cline	JSON	%APPDATA%\Code\User\globalStorage\saoudrizwan.claude-dev\...
Claude Code	JSON	~/.claude.json
Zed	JSON	~/.config/zed/settings.json
Continue	YAML	~/.continue/config.yaml
Goose	YAML	~/.config/goose/config.yaml

mcp-hub auto-detects which clients are installed and writes to all of them.

Continue and Goose require pyyaml:

pip install "mcp-hub[yaml]"

---

How it works

mcp-hub only writes JSON config files — it does not install packages directly. Servers are started lazily by your AI client using npx -y (Node.js) or uvx (Python), which handle downloading automatically.

This means:

• mcp install is near-instant — no network calls
• No version conflicts or global npm installs
• The server always runs the latest version

---

Requirements

• Python 3.10+
• Node.js (for npx-based servers) — nodejs.org
• uv (for uvx-based servers) — astral.sh/uv

---

License

MIT