Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/neon-solutions/add-mcp/llms.txt

Use this file to discover all available pages before exploring further.

When you run add-mcp without specifying an agent, it doesn’t ask you to pick from a blank list — it scans your environment first. This smart detection step finds which AI coding agents are active in your current directory (or on your machine globally) and pre-selects them in the interactive prompt, so the right agents are already checked before you press Enter.

​
How detection works

Detection logic differs based on whether you are installing in project scope (the default) or global scope (-g).
add-mcp scans the current directory for agent config markers. Each agent has a set of paths it looks for — for example, Cursor is detected when .cursor/ exists, Claude Code when .mcp.json or .claude/ is present, and VS Code / GitHub Copilot CLI when .vscode/ exists.Detected agents appear at the top of the multi-select list with a project hint and are pre-checked. You can deselect them or add others before confirming.Detected project markers by agent:
AgentDetection paths
Claude Code.mcp.json, .claude/
Codex.codex/
Cursor.cursor/
Gemini CLI.gemini/
GitHub Copilot CLI.vscode/
MCPorterconfig/mcporter.json
OpenCodeopencode.json, .opencode/
VS Code.vscode/
Zed.zed/
Agents that are global-only (Antigravity, Cline, Cline CLI, Claude Desktop, Goose, Windsurf) are never detected in project mode but still appear in the list so you can select them manually.

​
Last-selection memory

add-mcp remembers the agents you chose the last time you ran an install. This selection is persisted to ~/.config/add-mcp/config.json (respecting $XDG_CONFIG_HOME) and used as the default pre-selection the next time no agents are detected. In the multi-select prompt, previously chosen agents that were not detected in the current environment appear below detected agents with a · selected last time hint, making it easy to spot them.
The last-selection memory is used as a fallback only when no agents are detected. Detected agents always take priority and are pre-selected regardless of the saved state.

​
Bypassing detection explicitly

If you already know which agents you want, pass -a one or more times. This skips detection and the interactive prompt entirely: 
# Bypass detection — install directly to Cursor and Claude Code
npx add-mcp https://mcp.example.com/mcp -a cursor -a claude-code

# Same, globally, non-interactive
npx add-mcp https://mcp.example.com/mcp -a cursor -a claude-code -g -y

​
The --all flag

--all installs to every supported agent at once, completely bypassing both detection and the interactive prompt: 
# Install to all agents, project scope where supported
npx add-mcp mcp-server-github --all

# Install to all agents, globally, no prompts
npx add-mcp mcp-server-github --all -g -y
Use --all --gitignore when you want a broad install across all project-capable agents without committing any of the generated config files to version control.
npx add-mcp https://mcp.example.com/mcp --all --gitignore -y

​
Detection in the programmatic API

You can run the same detection logic from your own tooling using the exported functions: 
import { detectProjectAgents, detectGlobalAgents } from "add-mcp";

// Returns AgentType[] of agents with project config in /my/project
const projectAgents = detectProjectAgents("/my/project");

// Returns AgentType[] of globally installed agents
const globalAgents = await detectGlobalAgents();
Both functions return an array of AgentType strings (e.g. ["cursor", "claude-code"]) that you can pass directly to upsertServer or use for any other purpose in your own CLI integrations.

Build docs developers (and LLMs) love

Get started for freeTalk to us