Add solvethree MCP to Claude Desktop

Install Node.js 22 on the machine running Claude Desktop. Have your workspace API key and an HTTP or HTTPS proxy URL ready. The MCP server uses one proxy supplied through SOLVETHREE_PROXY; the reCAPTCHA tool doesn't accept a proxy argument.

For reCAPTCHA, first create an active profile in Site profiles. Configure the target hostname, capture action, and reCAPTCHA type you need. Supplying a hostname and action to Claude doesn't create that profile. See site profile setup if it isn't active yet.

1. 1.

   Open the local configuration file

   On macOS, use ~/Library/Application Support/Claude/claude_desktop_config.json. On Windows, use %APPDATA%/Claude/claude_desktop_config.json.

2. 2.

   Add the solvethree server

   Use the configuration below. If the file already contains mcpServers, add the solvethree entry inside it rather than replacing your other servers. Replace both environment-variable placeholders privately.

3. 3.

   Restart Claude Desktop

   Restart after saving so the host launches the server with the new configuration.

{
  "mcpServers": {
    "solvethree": {
      "command": "npx",
      "args": [
        "--yes",
        "--package=https://solvethree.com/downloads/solvethree-sdk-0.1.0.tgz",
        "solvethree-mcp"
      ],
      "env": {
        "SOLVETHREE_API_KEY": "REPLACE_WITH_YOUR_WORKSPACE_KEY",
        "SOLVETHREE_PROXY": "http://USER:PASSWORD@YOUR_PROXY:PORT"
      }
    }
  }
}

The args download the official SDK archive and launch its solvethree-mcp executable as a local stdio server. There is no hosted MCP endpoint to enter.

Start with solvethree_capabilities. Use solvethree_get_task to retrieve an existing task. Neither operation is a request to create a new solve.

• solvethree_create_recaptcha accepts site as a hostname, action, and requestId, with optional recaptchaType and waitMs. The default type is enterprise-v3; v3 is also accepted.
• solvethree_create_kasada creates a Kasada task. Standalone Kasada currently supports HTTPS targets on www.ticketmaster.com, not arbitrary websites.
• solvethree_create_tmpt creates a TMPT task. Event tasks use the event URL; Login tasks use the sign-in entry URL.

Give Claude explicit permission to create one task for the agreed target and action. Require fresh approval before it starts another paid task, including a replacement after failure or target rejection. Permission to inspect or poll a task shouldn't be treated as permission to create another.

Every create tool requires requestId and makes at most one task. Keep that ID with the original inputs. If creation times out, recover the task by repeating the same request with the same ID, or retrieve it if you already have its task ID. A timeout doesn't prove creation failed. Changing the inputs while reusing the ID conflicts; using a new ID can create another task.

For pending or solving, continue retrieving the existing task. Only ready contains a usable solution. On failed, stop polling and inspect errorCode and error; failed solves are refunded. Don't automatically submit the same invalid inputs again.

Claude receives private solution data through the MCP host. Have your application use the result for its intended target, with the successful proxy and returned userAgent and secChUa. Don't solve through the configured proxy and then submit through a different IP or unrelated browser identity.

A ready task means a solution was generated, not that the website accepted your subsequent request. If the target rejects it, inspect the outgoing token, cookie or headers, proxy, browser identity, and target URL before authorizing another solve.