MCP server managers for Claude Code, Codex and Gemini CLI (2026)

Short answer: If you run Claude Code, Codex and Gemini CLI in your own terminals and want one MCP server list for all of them, use a desktop manager that writes each client’s config: Toolport (MIT, with keychain secrets and paid sync across machines) or MCP Manager by xjeway (MIT, with a registry browser). Codeg, an open-source multi-agent workspace, does the same for the agents it runs and installs servers from the official MCP Registry and Smithery. If you run your agents inside one app, AgentVera’s MCP manager passes the same servers to Claude Code, Codex, Gemini CLI and 14 more CLI agents when they start, with one sign-in for remote servers.

We build AgentVera, one of the tools below. We have described every tool, ours included, from its own site, docs or repository as of 3 November 2026. If you only use Claude Code, our Claude Code MCP setup guide covers the built-in commands, and the best MCP servers for coding covers what to install.

Three ways to keep one MCP list

• Write every client’s config. The manager keeps the master list and writes it into ~/.claude.json, Codex’s config.toml, ~/.gemini/settings.json, ~/.cursor/mcp.json and so on. Codeg and xjeway’s MCP Manager work this way.
• Point every client at one hub. Each client gets a single entry that leads to a local gateway, and the gateway runs the real servers. Toolport and vietqtran’s MCP Manager work this way.
• Hand servers over at launch. The app that starts the agent passes the servers on the command line, so the client’s own file stays as it was. AgentVera works this way for the agents it runs.

Built-in settings in Claude Desktop, Cursor and VS Code are a fourth option, but each one manages only its own client.

MCP server managers compared

Checked on each tool’s own site, docs or repository on 3 November 2026.

Notes: MCP Router, another desktop manager, ended development on 18 September 2026 and its repository is archived, so it is not in the table. vietqtran’s MCP Manager has no published release yet; its README describes building from source. This space changes fast, so check each project before you decide.

Which one fits?

You run agents in your own terminals

Pick a manager that writes the configs, so the servers are there whichever terminal you open. Toolport suits you if you want a polished app with keys in the keychain and, later, the same setup on several machines or for a team. xjeway’s MCP Manager suits you if you want a free, MIT app that browses MCP registries and applies one list to 24 clients with a review step. vietqtran’s MCP Manager suits a Mac user who prefers one shared engine and one entry per client over copies in every file.

You want one workspace for the agents too

Codeg and AgentVera both run several coding agents side by side, so MCP management is part of a larger app. Codeg writes servers into each agent’s native config and installs from two public registries; it is open source and also runs as a server with mobile clients. AgentVera hands servers to each agent at launch and adds remote-server sign-in, token cost per tool and servers per agent group.

You use one client

Use its built-in settings. Claude Desktop, Cursor and VS Code all have a server list, and Cursor and VS Code have one-click installs. For Claude Code on its own, claude mcp add is enough; see our setup guide.

Where secrets live

If API keys in plain JSON or TOML files worry you, look for keychain storage (Toolport, AgentVera), a secrets manager (Toolport Pro) or environment variable references (Cursor). Codeg’s docs say plainly that keys written into agent configs are not encrypted at rest.

How AgentVera’s MCP manager works

AgentVera is a desktop app for macOS, Windows and Linux that runs coding agents side by side in real terminals. Its MCP manager keeps one server list for all of them:

• Add a server once. On the MCP page (⌘⌥M), “+” lists 23 ready servers by category, such as Playwright, Chrome DevTools, Context7, GitHub, Sentry, Supabase, Linear, Notion, Stripe and Figma. You can also add your own stdio, HTTP or SSE server.
• Every CLI agent gets it. Registered servers connect when an agent starts: Claude Code, Codex, Gemini CLI, opencode, Copilot, Qwen Code, Goose, Amp, Cursor, Antigravity, Droid, Kimi, Vibe, Crush, Kiro, Cline and Hermes. Aider takes no MCP. You choose all agents, some agents, a group or a team role.
• One sign-in for remote servers. Remote servers go through AgentVera’s gateway. You sign in to an OAuth server once in the app and every agent uses that sign-in; the agent’s CLI holds no header or token.
• Test it, see the cost. A connection test lists the server’s tools, prompts and resources and an approximate token cost per tool, so you know how much context a server adds. Secret environment variables and headers are kept in the operating system’s keychain, and servers can be limited to certain projects.
• In step with your CLIs. Servers from Claude Code’s and Codex’s own settings are listed with their health; import them or remove them from the CLI. AgentVera’s servers can be written to the CLI settings too.
• Agents ask for servers. An agent can attach a registered server to itself or to an agent of its group. A new server is only added when you approve it in the app.

What to weigh: AgentVera is not open source. The MCP manager hands servers to agents that AgentVera starts; a claude you open in another terminal sees only what is in its own settings, unless you write AgentVera’s servers there. The MCP manager is on the free plan; see pricing.

Originally published on the AgentVera blog.