Power your agent
Perintah milik penerbit sendiri. Bukan endpoint PlugMCP.
Host a copy with our keys and logs.
npx -y [email protected]Claude Code tidak punya tautan instal untuk perintah lokal. Salin perintah di bawah dan jalankan di terminal.
claude mcp add add-mcp -- npx -y add-mcp@2.6.1Jalankan di terminal.
{ "mcpServers": { "add-mcp": { "command": "npx", "args": [ "-y", "[email protected]" ] } }}claude_desktop_config.json
URL remote atau perintah di mesin Anda? Perbedaan kedua cara instal
Jalankan di PlugMCP
A new server you create. It is not this listing's official endpoint. Use the publisher install when you want their URL and their credentials.
BukaDari penerbit
Add MCP servers to your favorite coding agents with a single command.
Supports Claude Code, Codex, Cursor, OpenCode, VS Code, Grok Build and 19 more.
Docs and registry: add-mcp.com
Add MCP servers to your favorite coding agents with a single command.
Install an MCP server by remote URL or package name:
npx add-mcp url | package name [options]
Example installing the Context7 remote MCP server:
npx add-mcp https://mcp.context7.com/mcp
Example installing the Context7 remote MCP server via the programmatic API — great for integrating add-mcp in your own CLI tool:
import { upsertServer } from "add-mcp";
const result = upsertServer(
"cursor",
"context7",
{ type: "http", url: "https://mcp.context7.com/mcp" },
{ local: true },
);
console.log(result);
You can add env variables and arguments (stdio) and headers (remote) to the server config using the --env, --args and --header options. With ${VAR} placeholders, interactive installs prompt for each variable (omit skipped optional keys so empty strings are not written to config).
Find and install MCP servers from the add-mcp registry:
npx add-mcp find vercel
The first find/search run automatically saves the add-mcp registry to ~/.config/add-mcp/config.json (or $XDG_CONFIG_HOME/add-mcp/config.json). You can edit that file to replace the default registry, add the official Anthropic registry, or point at any registry-compatible server.
MCP servers can be installed to any of these agents:
Aliases: codeium, cascade → windsurf, cline-vscode → cline, gemini → gemini-cli, github-copilot → vscode, grok → grok-build, kilo, kilocode → kilo-code, kimi → kimi-code, kiro → kiro-cli, mastra → mastracode, pi-agent → pi
Mastra Code also reads project .mcp.json for Claude Code compatibility. add-mcp writes .mastracode/mcp.json so a mixed Claude + Mastra install does not share that file.
Pi has no built-in MCP. add-mcp writes the Pi-owned files pi-mcp-adapter reads (pi install npm:pi-mcp-adapter). A running Pi session picks the change up after /reload.
Kimi Code only loads a project-level .kimi-code/mcp.json after you trust the folder in the CLI, so a project install may not take effect until then.
Junie CLI and the Junie JetBrains IDE plugin read the same mcp.json. The IDE plugin always reads ~/.junie/mcp/mcp.json, so it only sees a global install when JUNIE_HOME is unset. Junie requires server names to match ^[a-zA-Z0-9_-]+$, so keep --name within that pattern. Junie CLI sends an Authorization header from the config; the IDE plugin does not support tokens in MCP configs.
Hermes Agent keeps MCP servers under mcp_servers in the same config.yaml as its other settings; add-mcp edits only that entry and keeps the file's comments. Without HERMES_HOME, add-mcp writes to the profile selected with hermes profile use. A running session picks the change up after /reload-mcp.
fx reads MCP servers only from ~/.fx/mcp.json. A project file cannot add a server. A running session applies the change with /mcp reload.
OpenCode keeps the shape already in the file: mcp.<name> on a V1 or empty config, mcp.servers.<name> when that native map is already there. In a mixed file, a native entry wins over a legacy entry with the same name. Adding a new name to a mixed file still writes V1; renaming a server keeps it in the map it already uses.
GitHub Copilot CLI project installs write .mcp.json. If .mcp.json is absent, an existing .github/mcp.json is reused. Copilot CLI does not read .vscode/mcp.json (that path is the VS Code agent). Entries left in .vscode/mcp.json from older Copilot CLI installs stay there for VS Code; rerun with --agent github-copilot-cli to write the CLI path. .mcp.json is shared with Claude Code: removing a server from either agent removes it for both. add-mcp will not create .mcp.json when .github/mcp.json already exists, because Copilot CLI would then ignore the GitHub file. Merge those servers into .mcp.json under mcpServers first. Copilot writes replace the file with strict JSON, so comments in that file are dropped (including global ~/.copilot/mcp-config.json). Copilot CLI loads a project file only after the folder is trusted; confirm with copilot mcp list --json.
The CLI automatically detects agents based on your environment:
Default (project mode):
.cursor/, .vscode/, .mcp.json, etc.)With -g (global mode):
No agents detected:
--yes: Installs to all project-capable agents (project mode) or all global-capable agents (global mode)Besides the implicit add command, add-mcp also supports the following commands:
# Remote MCP server (streamable HTTP)
npx add-mcp https://mcp.example.com/mcp
# Remote MCP server (SSE transport)
npx add-mcp https://mcp.example.com/sse --transport sse
# Remote MCP server with auth header
npx add-mcp https://mcp.example.com/mcp --header "Authorization: Bearer $TOKEN"
# Remote server; fx reads the bearer token from this env var
npx add-mcp https://mcp.example.com/mcp --bearer-token-env NEON_API_KEY
# Remote server with a request timeout and OAuth scopes
# (each agent only keeps the fields it supports; others are dropped with a warning)
npx add-mcp https://mcp.example.com/mcp --timeout 30000 --scopes "read,write"
# Auto-approve all tools for agents that support it (Codex, Claude Code)
npx add-mcp "executor mcp" --name executor -a codex -a claude-code --auto-approve
# Auto-approve only selected tools
npx add-mcp "executor mcp" --name executor -a codex --auto-approve --approve-tool execute
# npm package (runs via npx)
npx add-mcp @modelcontextprotocol/server-postgres
# Non-interactive installation to all detected agents in the project directory
npx add-mcp https://mcp.example.com/mcp -y
# Non-interactive installation to the global Claude Code config
npx add-mcp https://mcp.example.com/mcp -g -a claude-code -y
# Full command with arguments
npx add-mcp "npx -y @org/mcp-server --flag value"
# Node.js script
npx add-mcp "node /path/to/server.js --port 3000"
# Local stdio server with environment variables (repeatable)
npx add-mcp @modelcontextprotocol/server-filesystem --env "API_KEY=secret" --env "DATABASE_URL=postgres://localhost/app"
# Install for Cursor and Claude Code
npx add-mcp https://mcp.example.com/mcp -a cursor -a claude-code
# Install with custom server name
npx add-mcp @modelcontextprotocol/server-postgres --name postgres
# Install to all supported agents
npx add-mcp mcp-server-github --all
# Install to all agents, globally, without prompts
npx add-mcp mcp-server-github --all -g -y
# Add generated config files to .gitignore
npx add-mcp https://mcp.example.com/mcp -a cursor -y --gitignore
--timeout, --scopes, --bearer-token-env)Not every MCP client understands every field. add-mcp keeps one canonical
server config and each agent declares which optional fields it supports, mapping
them into that client's native shape:
When you target an agent that does not support a field, add-mcp drops it from
that agent's config and prints a warning (e.g. "request timeout is not
supported by VS Code; dropped from that config."). Other agents still receive
it. --timeout, --scopes, and --bearer-token-env apply to remote servers
only; --auto-approve applies to both remote and local servers.
When --bearer-token-env and --header Authorization are both set, fx omits
the Authorization header and writes bearer_token_env. Other agents keep the
header.
--auto-approve)--auto-approve preconfigures agent-level approval so the agent doesn't prompt
before each MCP tool call — useful for servers that already gate actions
internally. Use --approve-tool <name> (repeatable) to approve only specific
tools; without it, all tools are approved.
config.toml: per-tool
tools.<name>.approval_mode = "approve", or default_tools_approval_mode = "approve" for all tools..claude/settings.local.json for project installs, ~/.claude/settings.json
for global), e.g. mcp__<server>__<tool>, or mcp__<server> for all tools.
The MCP server entry itself stays clean.Note: Claude Code's all-tools rule (
mcp__<server>) follows the documented format, but a known Claude Code bug (#34739) can still prompt for non-fully-qualified MCP rules. Listing tools explicitly with--approve-toolis the most reliable path there.
add-mcp supports all three transport types: HTTP, SSE, and stdio. Some agents require type option to be set to specify the transport type. You can use the --type or --transport option to specify the transport type:
Local servers (npm packages, commands) always use stdio transport.
Note that most agents like Cursor and opencode do not require the type information to be set.
# Search for servers by keyword and choose one interactively
npx add-mcp find vercel
# Browse servers without a keyword
npx add-mcp find
# Use search alias (same as find)
npx add-mcp search notion
# Install a found server globally to a specific agent without prompts
npx add-mcp find neon -a claude-code -g -y
# Install to all agents and add generated project configs to .gitignore
npx add-mcp find github --all --gitignore
Transport for find/search is inferred from registry metadata. The CLI prefers HTTP remotes when available and only falls back to SSE when HTTP is not available for the selected install context.
When a server offers both remote and stdio package options, interactive mode lets you choose one (remote is the default). With -y, it auto-selects remote.
If a selected remote server defines URL variables or header inputs:
-y, placeholders are inserted (for example <your-header-value-here>)The first time you run find or search, the CLI automatically saves the add-mcp registry to ~/.config/add-mcp/config.json (respects XDG_CONFIG_HOME) and reuses that registry on every subsequent search.
The add-mcp registry previously lived at
https://mcp.agent-tooling.dev/api/v1/servers. That URL keeps working, and saved configs referencing it are migrated to the add-mcp.com address automatically on the nextfind/searchrun.
The default add-mcp registry is generated from integrations.sh. To be listed in add-mcp, add your MCP server to integrations.sh. Package-only servers, and remotes integrations.sh dropped that we still want in find, go in registry.overlay.json.
Maintainers can refresh the checked-in registry snapshot with:
bun run registry:sync
bun run registry:verify
Registry selections are stored in ~/.config/add-mcp/config.json under the findRegistries key. You can edit this file directly to replace, add, remove, or reorder registries.
Default add-mcp registry:
{
"version": 1,
"findRegistries": [
{
"url": "https://add-mcp.com/registry/api/v1/servers",
"label": "add-mcp registry"
}
]
}
Replace the default with the official Anthropic registry:
{
"version": 1,
"findRegistries": [
{
"url": "https://registry.modelcontextprotocol.io/v0.1/servers",
"label": "Official Anthropic registry"
}
]
}
Search both the add-mcp registry and the official Anthropic registry:
{
"version": 1,
"findRegistries": [
{
"url": "https://add-mcp.com/registry/api/v1/servers",
"label": "add-mcp registry"
},
{
"url": "https://registry.modelcontextprotocol.io/v0.1/servers",
"label": "Official Anthropic registry"
}
]
}
To reset to the default add-mcp registry, remove the findRegistries key or delete the config file.
Any server that implements the registry API can be added as a custom entry. The CLI sends a GET request to the configured url with the following query parameters:
The endpoint must return JSON in this shape:
{
"servers": [
{
"server": {
"name": "example-server",
"description": "An example MCP server",
"version": "1.0.0",
"remotes": [
{
"type": "streamable-http",
"url": "https://mcp.example.com/mcp"
}
]
}
}
]
}
To add your own registry, append an entry to findRegistries in ~/.config/add-mcp/config.json:
{
"url": "https://my-registry.example.com/api/v1/servers",
"label": "My custom registry"
}
List installed MCP servers across detected agents:
# List servers for all detected agents in the project
npx add-mcp list
# List global server configs
npx add-mcp list -g
# List servers for a specific agent (shown even if not detected)
npx add-mcp list -a cursor
Remove an MCP server from agent configurations by server name, URL, or package name:
# Remove by server name (interactive selection by default)
npx add-mcp remove neon
# Remove all matches without prompting
npx add-mcp remove neon -y
# Remove by URL
npx add-mcp remove https://mcp.neon.tech/mcp -y
# Remove from global configs for a specific agent
npx add-mcp remove neon -g -a cursor -y
Synchronize server names and installations across all detected agents. Servers are grouped by URL or package name, and each group is unified to the shortest server name. Servers with conflicting headers, env, or args across agents are skipped with a warning.
# Sync project-level configs (interactive confirmation)
npx add-mcp sync
# Sync without prompting
npx add-mcp sync -y
# Sync global configs
npx add-mcp sync -g -y
unify is an alias for sync.
import {
detectProjectAgents,
detectGlobalAgents,
upsertServer,
removeServer,
listInstalledServers,
} from "add-mcp";
const project = detectProjectAgents("/path/to/project");
const global = await detectGlobalAgents();
// Remote server
upsertServer(
"claude-code",
"example",
{ type: "http", url: "https://mcp.example.com/api" },
{ local: true, cwd: "/path/to/project" },
);
// Stdio (npm package)
upsertServer("claude-code", "postgres", {
command: "npx",
args: ["-y", "@modelcontextprotocol/server-postgres"],
});
const projectServers = await listInstalledServers({
cwd: "/path/to/project",
});
const globalServers = await listInstalledServers({ global: true });
removeServer("claude-code", "example", {
local: true,
cwd: "/path/to/project",
});
upsertServer and removeServer return { success, path, error? } rather than throwing.
Some agents & editors like Claude Code require a restart to load the new MCP server. Otherwise, like Cursor, require you to navigate to the MCP settings page and toggle the new server as enabled.
fx reads MCP servers only from ~/.fx/mcp.json. A project file cannot add a server. A running session applies the change with /mcp reload.
Add MCP servers to your favorite coding agents with a single command.
{ "mcpServers": { "add-mcp": { "command": "npx", "args": [ "-y", "[email protected]" ] } }}Tool
Belum ada daftar tool. Server remote mendapatkannya setelah pemeriksaan kesehatan pertama.
FAQ
Salin konfigurasi stdio untuk Cursor, Claude Desktop, Claude Code, atau VS Code di halaman ini. Konfigurasi ini memakai instalasi dari penerbit, bukan proxy PlugMCP.
Tidak ada variabel lingkungan wajib yang tercantum di catatan ini.
Add berjalan secara lokal melalui stdio.
Terakhir diperiksa 2026-11-13. Add MCP servers to your favorite coding agents with a single command.