MCP

MCP lets you connect Analyse to tools like Cursor, Claude, and other MCP clients. Instead of opening the dashboard for every small task, your AI client can read stats, list Servers and Networks, and perform dashboard-style management actions through controlled tools.

It is designed for operators who already trust their AI client with day-to-day server work, but still need clear boundaries around access.

::::info Plan Anyone can create MCP tokens. Tools can only act on Servers and Networks where the owner has Enterprise enabled. Creating new Servers or Networks through MCP also requires your account to have Enterprise, because you become the owner. ::::

What MCP can do

MCP tools are the programmatic version of dashboard actions. The client does not get a database console, SQL runner, Prisma access, or arbitrary query endpoint.

Today the MCP server exposes tools for:

• Discovery - list Servers, list Networks, open a Server or Network by ID
• Stats and reads - project stats, players, campaigns, creator codes, domains, members, dashboards, funnels, A/B tests, integrations, knowledge, report schedules, IP mappings
• Server and Network management - create, rename, delete, attach Servers to Networks, detach Servers from Networks
• Campaigns - create, update, delete
• Domains and IP mappings - add/delete domains and create/delete IP mappings
• Creator codes - create, update, delete
• Project API keys - create/delete plugin API keys

Destructive tools require an explicit confirm: true argument, even when the token is allowed to use the tool.

Security model

MCP access has four layers:

1. Bearer token - every MCP request uses a token created in your account settings.
2. Owner Enterprise gate - tools only work for Servers and Networks whose owner has Enterprise MCP access.
3. Existing permissions - team roles and permissions still apply. For example, a token for a viewer cannot manage domains.
4. Per-token tool access - in the dashboard you can disable individual tools for each token.

This means you can create a token for a specific workflow. For example:

• A read-only token for reporting agents
• A token that can edit campaigns but cannot delete anything
• A full-ops token for trusted internal automation on Enterprise-owned resources

::::warning Tokens are secrets Treat MCP tokens like passwords. Store them in environment variables when possible, revoke unused tokens, and create separate tokens for separate clients. ::::

Create an MCP token

1. 1

   Open Connections

   Go to Dashboard → Settings → Connections.

2. 2

   Open MCP

   Find Model Context Protocol and click Manage, or go directly to Dashboard → Settings → MCP.

3. 3

   Create a token

   Give it a clear name, such as Cursor - Wesley MacBook or Claude Desktop.

4. 4

   Copy the token

   The token is only shown once. If you lose it, revoke it and create a new one.

5. 5

   Configure tool access

   Select the token and enable or disable tools. Use the presets if you want a quick starting point:

   • Enable all
   • Read-only
   • No destructive

Add Analyse MCP to Cursor

Add Analyse to your Cursor MCP config. You can do this globally in ~/.cursor/mcp.json, or per project in .cursor/mcp.json.

{
  "mcpServers": {
    "analyse": {
      "url": "https://analyse.net/api/mcp",
      "headers": {
        "Authorization": "Bearer mcp_your_token_here"
      }
    }
  }
}

Replace mcp_your_token_here with the token you copied from Analyse. Keep this mcp.json out of git because it contains a secret.

Local development

When developing Analyse locally, run the web app:

pnpm --filter @analyse/web dev

Use the local endpoint:

{
  "mcpServers": {
    "analyse-local": {
      "url": "http://localhost:3000/api/mcp",
      "headers": {
        "Authorization": "Bearer mcp_your_token_here"
      }
    }
  }
}

After changing token tool access, refresh the MCP server in your client so it fetches the new tool list.

Disable tools for a token

Open Dashboard → Settings → MCP, select the Tool access tab, select a token, and toggle tools on or off.

Disabled tools are removed from that token's MCP tool list. The client will not see them and cannot call them.

Useful setups:

Read-only

Use this for analytics/reporting agents. It can list Servers, read stats, and inspect configuration, but it cannot create, update, or delete anything.

No destructive

Use this for operational helpers. They can create and update resources but cannot call tools like delete Server, delete Network, delete Campaign, or delete domain.

Custom

Turn tools on and off one by one. For example, you can allow updateNetwork so an agent can rename Networks, but disable deleteNetwork.

Troubleshooting

Cursor says no tools are available

Check that:

• the web app is running if you use localhost
• the MCP token is valid and not revoked
• the Server or Network owner has Enterprise
• your MCP client was refreshed after changing tool access

The agent cannot access a Server

The token uses your user permissions. Make sure your account can open that Server in the dashboard, and make sure the Server owner has Enterprise MCP access.

A tool exists but fails with a permission error

The token may see the tool, but the specific Server/Network action still checks role permissions. For example, manage_domains is required to add or delete domains.

I changed tool access but Cursor still sees the old tools

Refresh or restart the MCP server in Cursor. Tool lists are discovered by the client and may be cached until you reconnect.