The right account.
Every call.

Plug every MCP server and account into one endpoint. The right one answers for whatever repo you're in.

Your tools, connected.Routing example / change the profile below
Claude Cursor Codex distributorOne MCP endpoint Notion / work Team workspace Notion / personal Your own workspace Other servers GitHub, Linear, and more Claude · Cursor · Codex distributorOne MCP endpoint Notion / work Notion / personal
notion.search notion/work
1 endpoint to add 0 config edits after setup 2+ accounts per provider 100% of calls logged
Watch it

route.

Pick a project, make a call. The trace is the order the gateway actually runs: explicit override beats project config beats git remote beats your active profile.

A write aimed at production stops and waits for you. Try supabase.execute_sql from the work project.

Project
Call

Set up once. Forget it's there.

Claude has one config. Cursor has another. Codex has a third. Add a second Notion account and you're editing all three again. Move it into Distributor once and stop.

Project profile notion/work notion/personal .dstb.yamlSame tool. Selected account.
Profile: worknotion → notion/work
Projects pick their own accounts

A small file in the repo, or just its git remote, picks the profile. cd into your notes app and the same notion.search hits personal instead of work.

notion.search → notion/work
reason: git_remote
github.create_issue →
github/work
reason: project_config
supabase.execute_sql → held
Every call logged, with a reason
↗ Create your endpoint
in two minutes
Public data becomes read-only tools

A weather API, or that CSV a government agency publishes hourly. Point dstb add at it and the readable parts join your endpoint as tools. Writes stay off.

$ dstb add https://api.example.com/openapi.json
detected OpenAPI 3.1, 34 operations
created 34 read-only tools

Four steps. None of them hurt.

Ten minutes, mostly OAuth screens. After that your tools just call and the right account answers.

  1. Create your endpoint

    Sign up. You get one MCP URL and an API key. That pair is all any client needs.

  2. Connect your accounts

    Run dstb connect with a server URL and sign in through your browser. That's one connection. Run it again for the second account and both stay, each under its own name.

    $ dstb connect https://mcp.notion.com/mcp
    saved as notion/work

  3. Group into profiles

    A profile maps providers to accounts. Work gets one, personal gets another.

  4. Point your tools at it

    Add the endpoint to Claude, Cursor, Codex, whatever else speaks MCP. From then on you make changes in the Distributor dashboard and every client sees them on its next call. Those configs stay closed.

Bring the servers you already use.

Anything with an MCP URL connects. These are the usual suspects.

Notion GitHub Supabase Linear Slack Google Vercel Sentry + any MCP URL Notion GitHub Supabase Linear Slack Google Vercel Sentry + any MCP URL

FAQ

The questions people ask before adding it.

What changes in Claude, Cursor, or Codex?+−

One entry: your Distributor endpoint. That's the last time you open those configs. Prefer to keep everything on your machine? The CLI serves the same catalog over stdio, no cloud involved.

I have two Notion workspaces. How does it pick?+−

Each workspace is its own connection with its own name: notion/work, notion/personal. Your profile maps the provider to one of them per project. Inside the work repo, calls go to the work workspace. In your notes app, personal. Override by hand whenever you want.

What keeps a call from hitting prod by accident?+−

Environment tags. When routing is ambiguous and one of the options is prod, Distributor stops and asks instead of guessing. You can also require confirmation on every prod write, even an unambiguous one.

What counts as a public data source?+−

A documented API or a published dataset, like an hourly CSV export, exposed as read-only tools: no webpage scraping, no writes to the source.

Do I have to know server URLs?+−

No. The dashboard has a directory of the big ones (Notion, GitHub, Linear, Stripe and about twenty more), one click each. For anything else, paste the URL and Distributor figures out what it's talking to. Servers that need OAuth walk you through sign-in.

What do you store?+−

Your connections, your tool list, and a log of every call: tool, account, reason, duration. Results pass through to your client. They aren't kept.