Early access · Remote MCP over HTTPS

The USB hub for your MCP servers.

Connect your remote MCP servers once. Put the tools you want into MCP groups. Add one MCP endpoint to each coding agent. Choose its MCP group when you authorize.

Use your Google account to open your dashboard.

No tokens in the endpoint. Remote HTTPS servers only.

Sources

LinearOAuth or API key
DeepWikiPublic · no sign-in
OpenAI Developer DocsPublic · no sign-in
Custom MCPAny remote HTTPS

frontend-agent3 of 31 tools

linear.list_issues
openai_docs.search_openai_docs
deepwiki.ask_wiki_question

+ research, triage

One MCP endpointhttps://proxy.example.invalid/mcp

Coding agents

Claude Code
Cursor
Codex
T3 Code
any remote client

IllustrationFive services and one custom server connect to three MCP groups. Each agent uses one MCP endpoint and the MCP group chosen at authorization.

The problem

Every agent wants its own MCP setup.

Repeated config
Each coding agent has its own config file, its own login, and its own server list.
Too many tools
One server can add dozens of tools. Agents get slower and less focused when they see all of them.
Hard to change
To remove one tool, you edit every agent again.

24 setups to keep in sync

6 connections, 1 endpoint for every agent

Six servers and four agents. Without a hub, each agent needs each server. With a hub, you connect each server once.

How it works

Plug in once. Pick tools. Point your agents.

IllustrationThe examples in these steps are not live data.

Connect your servers
Pick Linear, DeepWiki, or OpenAI Developer Docs from the catalog, or add any remote MCP server URL. Linear authorizes access with OAuth or an API key, and credentials stay encrypted on the server. DeepWiki and OpenAI Developer Docs are public and need no sign-in.
- LinearOAuth or API key
- DeepWikiPublic · no sign-in
- OpenAI Developer DocsPublic · no sign-in
Build MCP groups
Make an MCP group for each job, for example frontend-agent or research. Approve only the tools that job needs. New or changed tools wait for your review.
research3 of 12 tools
- deepwiki.read_wiki_contents
- openai_docs.fetch_openai_doc
- custom.search_docs
Plug in one endpoint
Add the MCP endpoint to your coding agent. When it authorizes in your browser, you choose its MCP group. The endpoint contains no token.
https://proxy.example.invalid/mcp
- Claude Code
- Cursor
- Codex
- T3 Code

Control

One switch per tool.

Each MCP group has its own allowlist. Turn a tool off and every agent on that MCP group loses it at once. Revoke one client without touching the others.

frontend-agent · allowed toolsIllustration
Tool	Source	Status
`linear.list_issues`	Linear	Approved
`openai_docs.search_openai_docs`	OpenAI Developer Docs	Approved
`linear.delete_issue`	Linear	Off
`deepwiki.ask_wiki_question`	DeepWiki	Pending review

Spec sheet

What it supports

Which servers?: Remote MCP servers over HTTPS. Local stdio servers are not supported.
Which agents?: Clients that support remote MCP servers with browser authorization. Check your client's docs.
Secret in the endpoint?: No. Each client authorizes in the browser and gets one MCP group. You can revoke it later.
Available now?: Private alpha for invited accounts. Provider permissions and setup steps vary by provider.

The USB hub for your MCP servers.

Every agent wants its own MCP setup.

Repeated config

Too many tools

Hard to change

Plug in once. Pick tools. Point your agents.

Connect your servers

Build MCP groups

Plug in one endpoint

One switch per tool.

What it supports

Configure once. Use it in every agent.