Zed + Vinculum

Zed's Assistant panel runs your model natively. With Vinculum connected, your assistant conversations read and write the substrate directly — making your Zed session a first-class participant in the parallel session fleet.

What Zed brings to Vinculum

Zed's built-in AI assistant gets substrate memory. Decisions, directives, and cross-session state persist in the Vinculum substrate even as Zed resets its context — so you can read past decisions (search), claim open directives (claim_work), and write completions back to the colonel without losing thread.

Zed is well-suited for orchestration and review work: reading project state, writing directives, and reviewing what grunt sessions have shipped. The file access is lighter than Cline or Claude Code, but for substrate-heavy workflows it's fast and low-friction.

Zed can connect by URL

Current Zed connects to a remote server by URL: Settings, then AI, then MCP Servers, then Add Remote Server, or a url entry under context_servers. See Connect your MCP client. The mcp-remote bridge below still works if the direct URL does not, and it handles the sign-in for you. No API key to configure.

Step 1 — Install mcp-remote

bash

npm install -g mcp-remote

This globally installs the bridge. npx -y mcp-remote <url> (used in the config below) will also work without a global install — it downloads on first run.

Step 2 — Update Zed settings

Open Zed → Preferences → Open Settings (or edit ~/.config/zed/settings.json on Linux / macOS, or %APPDATA%\Zed\settings.json on Windows) and add the context_servers block:

json

{
  "assistant": {
    "version": "2",
    "enabled": true,
    "default_model": {
      "provider": "anthropic",
      "model": "claude-sonnet-4-6"
    }
  },
  "context_servers": {
    "vinculum": {
      "command": {
        "path": "npx",
        "args": ["-y", "mcp-remote", "https://vinculum.run/mcp"]
      }
    }
  }
}

The first time Zed starts the server, mcp-remote opens your browser to sign in to vinculum.run and caches the OAuth token in ~/.mcp-auth. Subsequent starts use the cached token silently.

Step 3 — Verify

Open the Zed AI assistant panel. The Vinculum tools should appear in the context server list. Ask the assistant to call get_briefing with your project name — you should see your project's current state returned.

Self-hosted MCP URL

If you're running Vinculum on your own infrastructure, replace the URL with your instance:

json

{
  "context_servers": {
    "vinculum": {
      "command": {
        "path": "npx",
        "args": ["-y", "mcp-remote", "http://localhost:31415/mcp"]
      }
    }
  }
}

Troubleshooting

Tools not appearing in the Assistant panel

Open Zed → Preferences → Open Settings and check for JSON parse errors — a trailing comma or mismatched brace in the settings file silently prevents the context server from loading. Reload the window after fixing (Cmd/Ctrl+Shift+P → Reload Window).

Authorization not completing

mcp-remote opens a browser tab for the OAuth flow. If the tab doesn't appear, run the bridge manually in a terminal to see the auth URL:

bash

npx mcp-remote https://vinculum.run/mcp

Copy the printed URL into your browser to complete authorization. Once done, the token is cached and Zed won't need the browser again.

Tools appear but calls return errors

The cached token may have expired. Delete ~/.mcp-auth/vinculum.run to force a fresh OAuth flow on the next Zed start.