Analysis updated 2026-05-18
Give Claude Code and Codex each their own curated set of MCP tools from one shared switchboard.
Manage OAuth logins and API keys for many MCP servers in a single place instead of per agent.
Run separate work and personal accounts of the same service as distinct, namespaced tool sets.
| cmer/mcp-switchboard | 000madz000/payload-test-api-route-handler | akash-sr/memobsidian | |
|---|---|---|---|
| Stars | 3 | 3 | 3 |
| Language | TypeScript | TypeScript | TypeScript |
| Setup difficulty | easy | easy | moderate |
| Complexity | 3/5 | 2/5 | 3/5 |
| Audience | developer | developer | general |
Figures from each repo's GitHub metadata at analysis time.
No built-in TLS or multi-user auth, intended for a trusted LAN, not the public internet.
MCP Switchboard is a small, self-hosted tool that acts as a central hub for MCP servers, the connectors that let AI coding agents like Claude Code or Codex reach outside tools and services. Instead of configuring each agent separately with its own set of servers, you register all your MCP servers once in the switchboard and then give each agent its own endpoint showing exactly the servers you want that agent to see. Setup is meant to be simple. Adding a popular service like GitHub, Linear, Notion, Stripe, or Supabase can be done through a visual template picker. Services that use OAuth send you straight to a browser consent screen, while others just ask for an API key. The switchboard also supports pasting an existing claude mcp add command or a raw mcpServers JSON block, which it parses and previews before creating the server entry. It handles both local servers, which run as processes it spawns and restarts if they crash, and remote servers reached over HTTP or SSE with various authentication methods, including keeping OAuth tokens refreshed automatically in the background. Each agent gets its own switch matrix, so you can turn individual servers on or off per agent without restarting anything, and changes apply live. Tools from different servers are namespaced, so two accounts of the same service, like a personal and a work Gmail, appear as separate, clearly labeled tool sets rather than colliding with each other. A built in tool lets agents themselves list which servers are connected and what they do. You can run it with a single npx command, install it globally, run it from source with hot reload for development, or deploy it with Docker, with state persisted to a local directory. Authentication is intentionally basic: one admin password for the web interface and one bearer token per agent, built for a trusted home network rather than the open internet. Secrets are encrypted at rest, but the tool has no built in TLS or multi-user support, so the README recommends putting something like Caddy or Tailscale in front of it for transport security.
A self-hosted gateway that centralizes your MCP servers and gives each coding agent its own endpoint with a per-agent switch for which servers it can see.
Mainly TypeScript. The stack also includes TypeScript, Node.js, SQLite.
Setup difficulty is rated easy, with roughly 5min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Verify against the repo before relying on details.