The problem: Every AI app uses a different config format for MCP servers. Claude Desktop uses mcpServers, VS Code uses servers, Goose uses YAML with cmd/envs, Codex uses TOML with mcp_servers... there are 19 apps, 6 root keys, and 4 formats.
The solution: One canonical format, config generators for every app, a registry of 105+ servers, and a CLI that auto-detects your apps and writes the correct config.
Browse the full server catalog at getmcp.es
Quick Start
# Install the GitHub MCP server into all your detected AI apps
npx @getmcp/cli add github
# Browse available servers
npx @getmcp/cli list
# Search for database servers
npx @getmcp/cli list --search=database
# Remove a server
npx @getmcp/cli remove github
Note: If you run into issues with npx, try npx @getmcp/cli@latest — npx can cache stale versions, and v0.7.0 had a broken release.
Why getmcp?
| Feature | getmcp | Smithery | mcpm.sh |
|---|
| Supported apps | 19 | 19 | 14 |
| Registry servers | 105+ | 100K+ (cloud) | 379 |
| Runtime dependency | None (write & done) | Cloud proxy | Local proxy |
| Config formats | JSON, JSONC, YAML, TOML | JSON only | JSON only |
| License | MIT | AGPL | MIT |
| Library-first design | Yes (npm packages) | No | No |
| Multi-app install | Yes (one command) | No | No |
| Offline/local-first | Yes | No | Partial |
How It Works
Canonical Format
|
+----------+-----------+-----------+----------+
| | | | |
Claude VS Code Goose Codex + 15 more
Desktop (servers) (YAML) (TOML) apps
(mcpServers) (extensions)(mcp_servers)
- Server definitions are stored in a canonical format (aligned with the official MCP registry schema)
- Config generators transform the canonical format into each app's specific format
- The CLI auto-detects installed apps, prompts for env vars, and merges configs into existing files (never overwrites)
Supported Apps
| App | Root Key | Format |
|---|
| Claude Desktop | mcpServers | JSON |
| Claude Code | mcpServers | JSON |
| VS Code / Copilot | servers | JSON |
| Cursor | mcpServers | JSON |
| Cline | mcpServers | JSON |
| Roo Code | mcpServers | JSON |
| Goose | extensions | YAML |
| Windsurf | mcpServers | JSON |
| OpenCode | mcp | JSONC |
| Zed | context_servers | JSON |
| PyCharm | mcpServers | JSON |
| Codex | mcp_servers | TOML |
| Gemini CLI | mcpServers | JSON |
| Continue | mcpServers | JSON |
| Amazon Q | mcpServers | JSON |
| Trae | mcpServers | JSON |
| BoltAI | mcpServers | JSON |
| LibreChat | mcpServers | YAML |
| Antigravity | mcpServers | JSON |
Registry
105+ MCP servers included out of the box. Here are some highlights:
| Server | Transport | Description |
|---|
| GitHub | stdio | Repository management, issues, PRs via GitHub API |
| Filesystem | stdio | Secure file operations with access controls |
| Brave Search | stdio | Web search via Brave Search API |
| Memory | stdio | Knowledge graph-based persistent memory |
| Playwright | stdio | Browser automation, screenshots, and element interaction |
| PostgreSQL | stdio | Read-only database access and queries |
| Figma | stdio | Read Figma design files and provide layout info to AI agents |
| shadcn/ui | stdio | Browse and install shadcn components from AI |
| Context7 | remote | Up-to-date library documentation and code examples |
| Sentry | remote | Error tracking and monitoring |
| Supabase | remote | Query databases, manage projects, deploy Edge Functions |
| OpenAI Docs | remote | Search OpenAI developer documentation |
| Firecrawl | stdio | Web scraping and search with JavaScript rendering |
| Repomix | stdio | Pack entire repositories into a single AI-friendly file |
| n8n | stdio | Build and manage n8n automation workflows from AI |
Browse the full catalog at getmcp.es or from the CLI:
npx @getmcp/cli list
Packages
Library Usage
Generate config for any app
import { generateConfig, generateAllConfigs } from "@getmcp/generators";
// Generate for a specific app
const config = generateConfig("goose", "github", {
command: "npx",
args: ["-y", "@modelcontextprotocol/server-github"],
env: { GITHUB_TOKEN: "ghp_xxx" },
});
// => { extensions: { github: { cmd: "npx", args: [...], envs: {...} } } }
// Generate for ALL apps at once
const all = generateAllConfigs("github", {
command: "npx",
args: ["-y", "@modelcontextprotocol/server-github"],
env: { GITHUB_TOKEN: "ghp_xxx" },
});
// all["claude-desktop"] => JSON string
// all["goose"] => YAML string
// all["codex"] => TOML string
Validate configs with Zod schemas
import { StdioServerConfig, CanonicalMCPConfig } from "@getmcp/core";
StdioServerConfig.parse({ command: "npx", args: ["server"] });
// throws ZodError if invalid
Search the registry
import { searchServers, getServersByCategory } from "@getmcp/registry";
searchServers("database"); // [{ id: "postgres", ... }]
getServersByCategory("web"); // [{ id: "brave-search", ... }, { id: "fetch", ... }]
Development
# Install dependencies
npm install
# Build all packages
npm run build
# Run all tests (515 tests)
npm run test
# Build and test a specific package
npm run build --workspace=@getmcp/core
npm run test --workspace=@getmcp/core
# Run the CLI locally (no build needed)
npx tsx packages/cli/src/bin.ts add
npx tsx packages/cli/src/bin.ts list --search=database
# Run the web directory locally
npm run website
Architecture
getmcp/
packages/
core/ # Zod schemas, TS types, utility functions
generators/ # 19 config generators (one per app)
registry/ # 105+ MCP server definitions
cli/ # add/remove/list/doctor/import commands, app detection
web/ # Next.js web directory at getmcp.es
The CLI supports JSON, JSONC, YAML, and TOML config files natively. Format is auto-detected from the file extension, so Goose configs are read/written as YAML and Codex configs as TOML.
Contributing
See ROADMAP.md for planned improvements and open tasks. Contributions are welcome.
Credits
This CLI is inspired by vercel-labs/skills.
License
MIT
