Skip to main content

Install the MCP

Viewfield runs a remote MCP server at https://viewfield.ai/api/mcp, authenticated with a Bearer token: your Viewfield API key, which scopes every call to your workspace.

First, create a key under Workspace settings › MCP (see Authentication). Then connect a client; the same tray shows a ready-made snippet for each.

The @viewfieldai/mcp package detects your client and wires everything up:

npx @viewfieldai/mcp install

It prompts for your API key, so the key never enters your shell history, and auto-detects installed clients: Claude Code (when the claude command is on your path), Cursor (~/.cursor), Claude Desktop (its config folder) and Windsurf (~/.codeium/windsurf). By default it also stores the key in ~/.reactor/config.json (readable only by you) for the bridge to use. Restart the client, then ask it to "list the Viewfield tools".

Target a specific client, or all of them:

npx @viewfieldai/mcp install --client cursor
npx @viewfieldai/mcp install --client all

In a script or a container, pass the key in the environment instead of on the command line:

REACTOR_API_KEY=sk_your_key npx @viewfieldai/mcp install --client cursor

Options:

  • --client claude-code|cursor|claude-desktop|windsurf|vscode|all (all writes every client except VS Code, which is written only when named, because its config is per project);
  • --project (write project-scoped config in the current directory, for Claude Code .mcp.json and Cursor .cursor/mcp.json, and VS Code's .vscode/mcp.json; those files then hold your key, so the installer offers to add them to .gitignore);
  • --key passes the key on the command line, which puts it in your shell history (the prompt avoids that);
  • --no-save skips storing the key in ~/.reactor/config.json.

npx @viewfieldai/mcp login stores a key for the bridge without touching any client config, and add is another name for install.

Config snippet​

If you would rather edit your client's config by hand, add the reactor entry (the server's name in a client config is kept for the clients already configured with it). The package is a small bridge over mcp-remote, so the endpoint stays out of your config:

{
  "mcpServers": {
    "reactor": {
      "command": "npx",
      "args": ["-y", "@viewfieldai/mcp"],
      "env": { "REACTOR_API_KEY": "YOUR_API_KEY" }
    }
  }
}
  • Claude Desktop: Settings › Developer › Edit Config (claude_desktop_config.json), add the entry, then restart.
  • Cursor: Settings › MCP › Add new MCP server, add the entry to mcp.json.
  • VS Code: a .vscode/mcp.json with a top-level servers block (npx @viewfieldai/mcp install --client vscode --project writes it for you).

npx -y @viewfieldai/mcp resolves the latest version at every editor restart; pin an exact version in args ("@viewfieldai/mcp@0.3.0") if you prefer.

Without the package​

The package wraps mcp-remote, which bridges the remote HTTP server to clients that speak local (stdio) MCP. You can call it directly:

npx -y mcp-remote https://viewfield.ai/api/mcp --header "Authorization: Bearer YOUR_API_KEY"

Clients that support remote HTTP MCP natively can skip the bridge: the MCP tray in Settings shows the entry with the endpoint's url and the Authorization header, for Cursor and other clients. In Claude Code:

claude mcp add reactor --transport http https://viewfield.ai/api/mcp --header "Authorization: Bearer YOUR_API_KEY"

Verify​

Ask your client to list the Viewfield tools, or call get_engine_config, which answers with your workspace's Brand. The server is self-describing; your client reads each tool's description and input schema over the protocol.