MCP Hub
Back to servers

Alph

Universal MCP Server Configuration Manager

GitHub
Stars
52
Forks
4
Updated
Oct 29, 2025
Validated
Jan 9, 2026

Alph banner

Alph — Configure MCP servers for your AI agents in one command. Local-first, atomic, and stress-free.

npm downloads GitHub stars MIT License

The Problem You Know Too Well

Every AI agent speaks a different configuration language. Cursor wants JSON in ~/.cursor/mcp.json. Claude expects it in ./.mcp.json. Gemini uses ~/.gemini/settings.json. One typo breaks everything. No backups. No validation. Manual editing is error-prone and time-consuming.

You've probably been there: copy-pasting server URLs, fixing bracket mismatches, and restarting your IDE hoping it works this time. What should take 30 seconds becomes a 30-minute debugging session.

Why Alph?

Modern AI agents speak MCP (Model Context Protocol), but wiring them up still means hand-editing fragile local config files per tool and per OS. Alph makes that painless: it detects installed agents, validates changes, performs atomic writes with timestamped backups, and gives you instant rollback — all without sending any network traffic.

Think of Alph as the universal remote for your AI developer tools. Point it at your MCP server (local or remote), pick your agent(s), and you’re done.


Demo

Quick Example

# Connect Cursor to your MCP server in one command
alph setup --mcp-server-endpoint https://api.example.com/mcp --bearer your-key --agents cursor

# ✅ Detects Cursor installation
# ✅ Validates configuration
# ✅ Creates timestamped backup
# ✅ Writes config atomically
# ✅ Verifies everything works

Before vs After

😰 Manual Way (Error-Prone)😌 Alph Way (Bulletproof)

Find config file: ~/.cursor/mcp.json

Edit by hand: Risk syntax errors

Manual restart: Hope it works 🤞

{
  "mcpServers": {
    "myserver": {
      "url": "https://api.example.com/mcp",
      "headers": {
        "Authorization": "Bearer sk-..."
      }
    }
  }
}

❌ No validation
❌ No backups
❌ Easy to break

One command: Works everywhere

alph setup \
  --mcp-server-endpoint https://api.example.com/mcp \
  --bearer sk-your-key \
  --agents cursor

✅ Auto-detects agents
✅ Creates backups
✅ Validates config
✅ Atomic writes
✅ Auto-rollback on error
Done! 🎉

Interactive Demo

Alph Demo

A quick wizard run: detect agents → choose transport → write configs → validate → done.

Try in 30 Seconds ⚡

# No installation needed - try it now
npx @aqualia/alph-cli@latest

# Or connect to your MCP server instantly:
npx @aqualia/alph-cli@latest setup \
  --mcp-server-endpoint https://your-server.com/mcp \
  --bearer your-api-key \
  --agents cursor,claude

Requirements: Node.js ≥ 18

Permanent Installation (if you like it)

# Global install for repeated use
npm install -g @aqualia/alph-cli

# Then just run:
alph

Supported AI Agents

Alph detects and configures these agents out of the box:

  • Gemini CLI (~/.gemini/settings.json)
  • Cursor
  • Claude Code
  • Windsurf
  • Codex CLI
  • Kiro (~/.kiro/settings/mcp.json)

Compatibility Matrix (OS × Transport)

AgentmacOSLinuxWindowsHTTPSSESTDIO
Gemini CLI
Cursor
Claude Code
Windsurf
Codex CLI
Kiro

MCP Transports 101 — Hosts/agents connect to servers via STDIO (local), HTTP, or SSE (streaming HTTP). Alph supports all three and lets you pick the best per agent.


Compatibility Notes

Codex CLI on Windows: Currently not supported due to upstream compatibility issues with Codex CLI's process spawning and environment variable handling on Windows. This is not an Alph-specific limitation but rather a known issue in the Codex CLI implementation. Multiple issues are actively tracked in the Codex repository, including #2555, #3311, and #3408. We recommend using Codex CLI on macOS or Linux for the best experience, or consider alternative agents like Cursor or Windsurf on Windows.


Usage

1) Interactive (recommended)

# Detects agents, guides transport choice, writes configs atomically with backups
alph
# or
alph setup
  • No flags needed: the wizard handles detection, preview, and safe writes.

2) Non-interactive (one-liner)

alph setup \
  --mcp-server-endpoint https://www.askhuman.net/api/mcp/YOUR_SERVER_ID \
  --bearer YOUR_ASKHUMAN_ACCESS_KEY \
  --agents gemini,cursor
  • Add --dry-run to preview changes without touching files.

3) STDIO tools (local servers)

For STDIO, Alph can auto-install the local tool (opt-out with --no-install) and run health checks before writing any config.

# Use STDIO without auto-installing the tool
alph setup --transport stdio --no-install

# Prefer a specific installer (npm|brew|pipx|cargo|auto)
alph setup --transport stdio --install-manager npm

# Control the atomic write strategy via env
ALPH_ATOMIC_MODE=copy alph setup --mcp-server-endpoint https://... --agents gemini

Quick Start (with AskHuman MCP server)

AskHuman is a remote MCP server. Use Alph to connect your local agents to your personal AskHuman endpoint in seconds.

# Configure AskHuman (remote MCP server) for multiple agents in one go
# Replace YOUR_SERVER_ID with the ID shown in AskHuman → Your MCP Server
alph setup \
  --mcp-server-endpoint https://www.askhuman.net/api/mcp/YOUR_SERVER_ID \
  --bearer YOUR_ASKHUMAN_ACCESS_KEY
  • Where to find these values: In the AskHuman dashboard (Your MCP Server), copy your endpoint (.../api/mcp/<your-id>) and generate an access key if auth is enabled.
  • Works with HTTP and SSE transports; choose --transport sse and use the .../api/mcp/<id>/sse endpoint if your host/agent prefers streams.

Safety & Reliability by Default

  • Local-first: Alph operates purely on local files; no network requests during configure/remove.
  • Atomic writes with timestamped backups and validation; fast rollback if anything looks off.
  • Secrets redacted in output (e.g., bearer tokens).

Command Reference (concise)

alph setup [options]
  --mcp-server-endpoint <url>  MCP server endpoint URL
  --bearer [token]             Authorization bearer token (optional)
  --transport <type>           http | sse | stdio
  --command <cmd>              Command for stdio transports
  --cwd <path>                 Working directory for stdio command
  --args <list>                Comma-separated args for stdio command
  --env <list>                 Env vars (KEY=VALUE)
  --headers <list>             HTTP headers (KEY=VALUE)
  --timeout <ms>               Command timeout
  --install-manager <mgr>      npm | brew | pipx | cargo | auto
  --atomic-mode <mode>         auto | copy | rename
  --no-install                 Skip auto-install for STDIO tools
  --agents <list>              Filter (e.g. gemini,cursor)
  --dir <path>                 Custom config root
  --dry-run                    Preview changes (no writes)
  --name <id>                  Name/ID for the MCP server

alph status [options]          Show detected agents and configured servers
  --dir <path>                 Include project-level configs where applicable (e.g., Claude)

alph remove [options]
      --server-name <name>         MCP server name to remove
      --agents <list>              Filter agents to modify
      --dir <path>                 Custom config root
      --scope <auto|global|project|all>  Scope for removal where supported (e.g., Claude)
      --dry-run                    Preview removal (no writes)
      -y, --yes                    Skip confirmation
      -i, --interactive            Removal wizard
      --no-backup                  Do not back up before removal (advanced)

Troubleshooting (quick hits)

  • Nothing seems to change → Re-run with --dry-run to confirm detection and planned writes; then run without it.
  • Agent not detected → Ensure the agent is installed and its config is in the default location (see agent docs in docs/agents).
  • STDIO tool missing → Add --install-manager npm (or brew|pipx|cargo) or run with --no-install if you want to manage it yourself.

Documentation

  • User GuideUSER_GUIDE.md (advanced setups, more recipes)
  • ArchitectureARCHITECTURE.md (execution flows & structure)
  • SecuritySECURITY.md (secret handling, backups, rollback)
  • TroubleshootingTROUBLESHOOTING.md
  • ContributingCONTRIBUTING.md

Community & Support

License

MIT — see LICENSE.


Reviews

No reviews yet

Sign in to write a review