Documentation · MCP

Connect Serebro to your AI app.

A step-by-step guide to expose your company’s brain through MCP — directly inside Claude Desktop, Cursor, Cline, and any MCP-compatible client.

~5 minutes Updated May 2026 Serebro MCP v1
What it is

Your company’s brain, one query away.

Serebro speaks the Model Context Protocol. Once configured, your AI client connects directly to your organizational graph — processes, decisions, project history, ticket trails, and every piece of context Serebro has harvested.

No copy-pasting. No retrieval gymnastics. Ask your assistant a question; the answer comes from your company’s actual state.

Note

Serebro keeps the graph current via Harvest. You don’t need to reconfigure when your knowledge changes — only when you rotate credentials, switch machines, or reinstall the client.

Before you start

What you’ll need.

01

Active Serebro workspace

Your team’s IAM-scoped access token. We show how to retrieve it in Step 1.

02

MCP-compatible client

Claude Desktop is the primary path. Cursor, Cline, Continue.dev and Zed also work.

03

Node.js installed

The MCP transport runs via npx. Install Node.js LTS from nodejs.org if it’s missing.

Step 01

Generate your IAM access token.

Tokens are scoped to your IAM permissions in Serebro. Each token belongs to one user and inherits their access. Treat them like passwords.

  1. Sign in to your Serebro portal.

    Navigate to app.serebro.io and log in with your workspace credentials.

  2. Open Settings → API & MCP → Access tokens.

    The settings menu lives in the top-right user dropdown.

  3. Click Generate new token.

    Give the token a name that identifies the machine — e.g., “Joel — MacBook Pro”.

  4. Copy the token immediately.

    Serebro shows the value once. You cannot retrieve it again. If lost, revoke it and generate a new one.

Heads up

Rotate tokens every 90 days. Revoke any time from the same screen — Serebro invalidates the credential in real time and your client disconnects on the next request.

Step 02

Open the Claude Desktop config.

Claude Desktop reads MCP servers from claude_desktop_config.json. Open it from inside Claude or navigate directly:

~/Library/Application Support/Claude/claude_desktop_config.json
%APPDATA%\Claude\claude_desktop_config.json
~/.config/Claude/claude_desktop_config.json
Tip

You can also reach the file from inside Claude: Settings → Developer → Edit Config.

Step 03

Drop in the Serebro server.

Paste the block below into your config file. If the file already has an mcpServers object, merge serebro into it. Replace YOUR_TOKEN_HERE with the value from Step 1.

claude_desktop_config.json
{
  "mcpServers": {
    "serebro": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://mcp.serebro.io/sse",
        "--header",
        "Authorization: Bearer YOUR_TOKEN_HERE"
      ]
    }
  }
}

Save the file once the block is in place.

Step 04

Restart Claude and confirm the connection.

  1. Quit Claude Desktop completely.

    macOS: ⌘Q. Windows: right-click the tray icon → Quit.

  2. Reopen Claude.

    The first launch after the config change will spawn the MCP process via npx; it may take a few seconds.

  3. Look for the tools icon next to the input field.

    A wrench / tool icon means MCP servers are active and Claude has tools available.

  4. Run a smoke test.

    Ask: “What did our team decide about [recent topic]?” Claude should answer with citations from your Serebro graph.

Other MCP clients

Same brain. Different clients.

Serebro’s MCP endpoint works with any MCP-compatible client. Pick yours below for the config path and snippet.

~/.cursor/mcp.json
~/.cursor/mcp.json
{
  "mcpServers": {
    "serebro": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://mcp.serebro.io/sse", "--header", "Authorization: Bearer YOUR_TOKEN_HERE"]
    }
  }
}

In VS Code: Cline → MCP Servers → Edit. Or edit the file directly:

~/Library/Application Support/Code/User/globalStorage/saoudrizwan.claude-dev/settings/cline_mcp_settings.json
cline_mcp_settings.json
{
  "mcpServers": {
    "serebro": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://mcp.serebro.io/sse", "--header", "Authorization: Bearer YOUR_TOKEN_HERE"]
    }
  }
}
~/.continue/config.json
~/.continue/config.json
{
  "experimental": {
    "modelContextProtocolServers": [
      {
        "transport": {
          "type": "stdio",
          "command": "npx",
          "args": ["-y", "mcp-remote", "https://mcp.serebro.io/sse", "--header", "Authorization: Bearer YOUR_TOKEN_HERE"]
        }
      }
    ]
  }
}

Edit Zed settings: ⌘, on macOS or Ctrl+, on Linux.

~/.config/zed/settings.json
{
  "context_servers": {
    "serebro": {
      "command": {
        "path": "npx",
        "args": ["-y", "mcp-remote", "https://mcp.serebro.io/sse", "--header", "Authorization: Bearer YOUR_TOKEN_HERE"]
      }
    }
  }
}
Troubleshooting

Common issues.

  • Server not detected after restart

    Node.js is likely missing. Verify with node --version. Install LTS from nodejs.org and relaunch your client.

  • Authentication failed

    Your token is invalid or revoked. Regenerate it in Settings → API & MCP and replace the value in your config.

  • Server listed but no tools appear

    JSON syntax error in the config. Validate at jsonlint.com — look for trailing commas or unmatched braces.

  • Connection times out

    Check that your network can reach mcp.serebro.io on port 443. A firewall or corporate proxy may be blocking the SSE connection.

  • Still stuck?

    Email manuel@omna.club with a screenshot of Settings → Developer in your client.

Automatic updates

The brain stays current.

Serebro continuously ingests from your connected tools through Harvest. The graph reflects the latest state of every source in real time, so your AI client always queries fresh data.

You only need to touch the config when you change machines, reinstall the client, or rotate your token.

Security

Token hygiene.

  • Tokens carry your IAM scope. Treat them like passwords.
  • Rotate every 90 days. Revoke any time from your portal — Serebro invalidates the credential immediately.
  • Never commit tokens to repos. Never share over email or Slack.
  • One token per machine. Don’t reuse across devices.

Need help wiring it up?

Reach the team on WhatsApp or email — we’ll walk you through it.